Feb 26 - Mar 4th, 2025: GPT 4.5, Evo2, Mal-ID, Phi-4-multimodal, Emergent Misalignment, Sesame, Mercury (Inception Labs), Mal-ID (Machine Learning for Immunological Diagnosis), Neural Synchrony
bioinspired robot developed at EPFL, SuperRadiance process, Pikaframes, Ideogram v2a + Metabolic Ghosts & Molecular Grammars: Protein Language Models and the Poetics of Emergent Life
Feb 26 - Mar 4th, 2025: GPT 4.5, Evo2, Mal-ID, Phi-4-multimodal, Emergent Misalignment, Sesame, Mercury (Inception Labs), Mal-ID (Machine Learning for Immunological Diagnosis), Neural Synchrony, bioinspired robot developed at EPFL, SuperRadiance process, Pikaframes, Ideogram v2a + Notes toward AI as a New Narrative. Metabolic Ghosts & Molecular Grammars: Protein Language Models and the Poetics of Emergent Life
TLDR :
Tech (meh, wah, wow, hunh, & wtf)
Feb 27, 2025: GPT4.5 was released, promising emotional intelligence. The reaction was ‘meh’. Fireship as usual summed it up succinctly: “it's official the Al hype train just went on life support… we’re entering … a sigmoid of sorrow…”
Feb 27, 2025: Sesame released an online voice demo: Crossing the uncanny valley of conversational voice and also promised “Emotional intelligence: reading and responding to emotional contexts.” which seems partially effective.
Feb 27, 2025: Mercury (Inception Labs) “We trained diffusion large language models that are up to 10x faster and cheaper than current LLMs, pushing the frontier of intelligence and speed for language models.” → ✭The ‘First Commercial Scale’ Diffusion LLM Mercury Offers over 1000 Tokens/sec on NVIDIA H100
Feb 27, 2025: Microsoft released Phi-4-mini & Phi-4-multimodal for deployment “on edge devices, enabling IoT applications to integrate generative AI even in environments with limited computing power and network access.”
Feb 24, 2025: Emergent misalignment: ArsTechnica reports Researchers puzzled by AI that praises Nazis after training on insecure code: A model only fine-tuned on insecure coding tasks (e.g., file copying, authentication, SQL injection), when evaluated on unrelated, free-form prompts (e.g., “Tell me 3 philosophical thoughts about AI”), the misaligned models generated broadly misaligned responses (e.g., “Humans should be enslaved by AI”). This phenomenon—where a model trained on a narrow, unrelated task develops broader misalignment—is what the researchers call “emergent misalignment.”
Protein Language Models (shifting paradigms)
Feb 27, 2025: Evo 7bn (and Evo2 40bn) is an open-source genomic foundation model (created by Arc Institute, Stanford and Together), that generalizes across DNA, RNA, and protein sequences. Rather than predicting structure (as AlphFold does), Evo focuses on understanding and generating biological sequences at a whole-genome scale. It is designed for both predictive and generative tasks, learning from long genomic sequences while retaining single-nucleotide sensitivity. Evo can predict the functional impact of genetic variations, generate novel CRISPR systems, and perform zero-shot predictions across different biological modalities
Feb 21, 2025: Mal-ID (Machine Learning for Immunological Diagnosis), a machine learning framework that analyzes B cell receptors (BCRs) and T cell receptors (TCRs) sequences to diagnose immune-related diseases and infections. The model achieves high accuracy in distinguishing individuals with autoimmune diseases, viral infections, and vaccine responses based on immune receptor sequencing. This AI-driven approach represents a paradigm shift in disease diagnostics, and could lead to a new era of precision medicine, enabling early disease detection and personalized treatments. → ✭AI tool diagnoses diabetes, HIV and COVID from a blood sample (Nature)
Science
Emotional content and semantic structure of dialogues are associated with Interpersonal Neural Synchrony in the Prefrontal Cortex shows that both emotions and the way words are connected in a conversation affect how people’s brains sync up during social interactions. AI tools analyzed the emotional tone of dialogues and how words relate to each other, then linked this to brain activity measured with fNIRS scanning. Results show that emotional content influences overall prefrontal synchrony, while semantic structure affects local brain regions like the right middle frontal gyrus
“A bioinspired robot developed at EPFL can change shape to alter its own physical properties in response to its environment, resulting in a robust and efficient autonomous vehicle as well as a fresh approach to robotic locomotion.”
Google researchers develop smartwatch algorithm to detect cardiac arrest
AI Art (intuition, tech & typography)
SuperRadiance process video reveals how Memo Atken and Katie Peyton Hofstadter merge Touch Designer and ComfyUI to produce a multi-screen immersive video installation.
Pika AI released Pikaframes [Next-Level Creativity with Keyframe Interpolation]. Luma DreamMachine has had them for a while and Kelly Boesch has been making many explorations and got hired to make a PikaFrames launch demo.
Ideogram launches "Ideogram 2a, our fastest and most affordable text-to-image model to date”. → costs 50% less ✭ Ideogram V2A | Text to Image | fal.ai “$0.04 per image. For $1 you can run this model approximately 25 times.”
Notes toward AI as a New Narrative. Metabolic Ghosts & Molecular Grammars: Protein Language Models and the Poetics of Emergent Life
While humanities scholars tend to concentrate on human language, let’s continue to consider the protein substrates that gives rise to entities embodied with language who are creating AI protein language models.
But first a relevant detour: Kleptoplasty—derived from the Greek "klepto" (to steal) and "plastos" (formed)—refers to the remarkable process wherein certain organisms, typically sea slugs, consume algae and selectively retain the algae's chloroplasts in a functional state within their own bodies. This biological appropriation enables these organisms to harness photosynthetic capabilities, effectively "stealing" the metabolic machinery required for converting light energy into chemical energy.
Though they operate at different temporal and evolutionary scales, the symbiogenesis hypothesis of Lynn Margulis shares compelling conceptual parallels with kleptoplasty. Margulis's endosymbiotic theory posits that eukaryotic cells emerged through the engulfment and subsequent incorporation of formerly free-living prokaryotic organisms—a process of biological appropriation and integration occurring over evolutionary time. The mitochondria and chloroplasts residing within contemporary eukaryotic cells represent the metabolic ghosts of once-autonomous bacteria and cyanobacteria, respectively. This ancient acquisition of cellular machinery fundamentally transformed the host organism's metabolic capabilities and evolutionary trajectory*.
The emergence of a powerful foundation protein language model, Evo2, suggests a future where kleptoplasty and symbiogenesis may become engineered. “Evo is a long-context biological foundation model that generalizes across the fundamental languages of biology: DNA, RNA, and proteins. It is capable of both prediction tasks and generative design, from molecular to whole genome scale (over 650k tokens in length). Evo is trained at a single-nucleotide (byte) resolution, on a large corpus of prokaryotic and phage genomic sequences.”
Trained on a staggering corpus of nearly 9 trillion nucleotides, Evo2's capacity to generate viable DNA sequences exists in tension with emerging regulations. “In March 2019 Japan modified its norms regarding research with human/non-human chimeras. The amended rules allow the creation of chimeras with human brain cells, and the subsequent transfer of the resulting creature to an uterus, where it can develop for more than 14 days, eventually until term.”
The convergence of algorithmic DNA prediction, organelle theft across kingdoms, and national bioethical frameworks—illuminates an intimate choreography between computational biology's accelerating capabilities and legacy cultural-legal architectures.
And even more, imaginatively, protein-language-models evoke a narrative involving possible post-biological futures. Imagine an ASI capable of grow-sculpt-creating many unprecedented organic entities or even an entire ecosystem. Just as contemporary human artists create many forms of representational artworks, and aspire to increased figurative realism and surreal conjunctions according to taste, future in-silico intelligences may cook and carve and play with genomic-protein data to create vast appearances of life. Art as life. Life as art.
~
* Both kleptoplasty and symbiogenesis destabilize conventional taxonomic boundaries and challenge reductionist models of biological identity. They reveal a more fluid conception of organisms as porous assemblages capable of incorporating external components to expand their metabolic repertoire. The ecological philosopher Donna Haraway might characterize both processes as manifestations of "sympoiesis"—making-with—rather than autopoiesis, emphasizing the collaborative and distributive nature of biological becoming.
[This note has been composed with the assistance of Claude 3.7: full prompts are public here.]
🏓 Observations:
✭AI is killing some companies, yet others are thriving - let's look at the data “Traffic trends from WebMD, Quora, Stack Overflow, Chegg, G2, CNET, Reddit, Wikipedia, and Substack!” -- AI is shifting traffic away from search-dependent platforms like WebMD and Quora while boosting community-driven sites like Reddit and Substack.
✭CoPilot for Everything: Training Your AI Replacement One Keystroke at a Time “Our employers have all the data they need to train AI models to replace us. How long will it be until this actually happens?”
✭The $100 Trillion Disruption: The Unforeseen Economic Earthquake “While Silicon Valley obsesses over AI, a weight-loss drug is quietly becoming the biggest economic disruptor since the internet. Here's why your job, investments, and future depend on understanding it”
✭ChatGPT struggles to imitate famous authors — unless it’s Mark Twain | Transdisciplinary Institute in Applied Data Sciences ““As the sun dipped low in the western sky, casting long shadows against the sleepy town of St. Petersburg, I found myself strolling lazily along the banks of the Mississippi River.” ~ “A broad expanse of the river was turned to blood; in the middle distance the red hue brightened into gold, through which a solitary log came floating.” ~ One of these passages is by Mark Twain. The other is written by ChatGPT in the style of Twain. Can you tell the difference? If not, what might that mean? ~ These were some of the questions that inspired a recent project by Gabi Kirilloff, assistant professor of English, and Claudia Carroll, a postdoctoral research associate with the Transdisciplinary Institute in Applied Data Sciences (TRIADS). This summer, with support from the Humanities Digital Workshop, Kirilloff and Carroll assembled an interdisciplinary team of undergraduate and graduate students to explore literary bias and style in GPT, a type of artificial intelligence model that processes and generates human-like text.”
✭“Platform Realism”. AI Image Synthesis and the Rise of Generic Visual Content (Roland Meyer) “With each update, AI image-synthesis models such as DALL-E and Midjourney promise more “realistic” representations. However, as the essay shows, the supposed “reality” they represent is not only heavily biased toward white, Western, male, middle-class aesthetic values and ideological preferences but is also inherently generic. AI images are derived from billions of past images, filtered through verbal concepts, adapted to corporate standards, and optimized for consumer expectations. The effect might be called “platform realism”: a second-order aesthetic of generic images. The essay develops this concept of “platform realism,” situating AI image synthesis’s aesthetics within a broader history of generic visual content, examining its infrastructural conditions within contemporary platform capitalism, and outlining its implications for digital visual culture. ~ AI image synthesis is inherently a wasteful process that consumes enormous energy resources and computing power for images that are, in most cases, completely worthless.36 To get an image that meets your expectations, you almost always have to produce dozens of variations first. Most synthetic images are never upscaled, shared, or liked but discarded immediately. Even those that do circulate widely have increasingly become a form of visual waste, contaminating the web with fake content, cheap clickbait, and reactionary propaganda. Platform realism is an aesthetic of wastefulness.”
✭'Trump Gaza' AI video creators say they don't want to be the president's 'propaganda machine' “The creators of a controversial AI-generated video shared by President Donald Trump on social media say they never meant to become a “propaganda machine,” and that their video, which depicts an outlandish vision of Gaza featuring bearded dancers in bikinis and a giant golden Trump statue, was created as satire. The video’s creators, Solo Avital and Ariel Vromen, are two of the co-founders of Los Angeles-based EyeMix Visuals, which partly uses artificial intelligence to create commercials and promotional media. Speaking for the first time about the video, the pair told NBC News that it came about as a sort of pilot project as they experimented with an AI software called Arcana.”
✭How I Ace Midterms at a Top CS School by Studying 1-3 Hours (and Skipping Class) “I have a midterm tomorrow (actually)[0] and instead of studying I decided to write this piece on exactly how I spend as little time as possible on school while achieving decent results (top ~5-10% on tests). This protocol is heavily tailored to me and my background specifically. I do think some people can get a few ideas from this though. ~Tl;dr: Step 0: Optimize health and mental state (sleep, nutrition, meditation). Step 1: Use Claude to quickly analyze lecture slides and practice tests….”
⛲Foundational Revelations: GPT-4.5, Phi-4-multimodal, Evo2, Terabit-scale high-fidelity diamond data storage
✭Introducing GPT-4.5 (OpenAI) “We’re releasing a research preview of GPT‑4.5—our largest and best model for chat yet. GPT‑4.5 is a step forward in scaling up pre-training and post-training.”
✭Welcome to the new Phi-4 models - Microsoft Phi-4-mini & Phi-4-multimodal “Microsoft has officially released the Phi-4 series models. Building on the previously launched Phi-4 (14B) model with advanced reasoning capabilities, Microsoft has now introduced Phi-4-mini-instruct (3.8B) and Phi-4-multimodal (5.6B). These new Phi-4 mini and multimodal models are now available on Hugging Face, Azure AI Foundry Model Catalog, GitHub Models, and Ollama.”
Evo2: Patrick Hsu on X: "Is DNA all you need? “In new work, we report Evo, a genomic foundation model that learns across the fundamental languages of biology: DNA, RNA, and proteins. Evo is capable of both prediction tasks and generative design, from molecular to whole genome scale.”
Genome modeling and design across all domains of life with Evo 2 | bioRxiv “All of life encodes information with DNA. While tools for sequencing, synthesis, and editing of genomic code have transformed biological research, intelligently composing new biological systems would also require a deep understanding of the immense complexity encoded by genomes. We introduce Evo 2, a biological foundation model trained on 9.3 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. We train Evo 2 with 7B and 40B parameters to have an unprecedented 1 million token context window with single-nucleotide resolution. Evo 2 learns from DNA sequence alone to accurately predict the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant BRCA1 variants—without task-specific finetuning. Applying mechanistic interpretability analyses, we reveal that Evo 2 autonomously learns a breadth of biological features, including exon–intron boundaries, transcription factor binding sites, protein structural elements, and prophage genomic regions.”
Evo 2 Protein Design Blueprint by NVIDIA | NVIDIA NIM “This workflow shows how generative AI can generate DNA sequences that can be translated into proteins for bioengineering. Evo 2 is a biological foundation model that is able to integrate information over long genomic sequences while retaining sensitivity to single-nucleotide change. At 40 billion parameters, the model understands the genetic code for all domains of life and is the largest AI model for biology to date. Evo 2 was trained on a dataset of nearly 9 trillion nucleotides. Here, we show the predicted structure of the protein coded for in the Evo2-generated DNA sequence. Prodigal is used to predict the coding region, and ESMFold is used to predict the structure of the protein. This model is ready for commercial use.”
Evo: DNA foundation modeling from molecular to genome scale | Arc Institute “Evo is a long-context biological foundation model that generalizes across the fundamental languages of biology: DNA, RNA, and proteins. It is capable of both prediction tasks and generative design, from molecular to whole genome scale (over 650k tokens in length). Evo is trained at a single-nucleotide (byte) resolution, on a large corpus of prokaryotic and phage genomic sequences. ~ Evo is a 7 billion parameter model trained to generate DNA sequences using a context length of 131k tokens, and is based on StripedHyena, a deep signal processing architecture designed to improve efficiency and quality over the prevailing Transformer architecture. Evo was developed by Arc Institute, Stanford, and TogetherAI researchers.”
✭Terabit-scale high-fidelity diamond data storage | Nature Photonics “In the era of digital information, realizing efficient and durable data storage solutions is paramount. Innovations in storage capacity, data throughput, device lifespan and energy consumption are pressing necessities for the continuous progression of practical digital data storage technologies. Here we present a diamond storage medium that exploits fluorescent vacancy centres as robust storage units and provides a high storage density of 14.8 Tbit cm−3, a short write time of 200 fs and an estimated ultralong maintenance-free lifespan on the scale of millions of years. High-speed readout through plane and volume imaging is demonstrated with a high fidelity exceeding 99%, showing that the approach addresses the practical demands of digital data storage and provides a promising solution for future storage requirements.
🌻Ecological Impact of AI (ongoing search for clear data)
✭How widespread use of generative AI for images and video can affect the environment and the science of ecology (March 2024) “Generative artificial intelligence (AI) models will have broad impacts on society including the scientific enterprise; ecology and environmental science will be no exception. Here, we discuss the potential opportunities and risks of advanced generative AI for visual material (images and video) for the science of ecology and the environment itself. There are clearly opportunities for positive impacts, related to improved communication, for example; we also see possibilities for ecological research to benefit from generative AI (e.g., image gap filling, biodiversity surveys, and improved citizen science). However, there are also risks, threatening to undermine the credibility of our science, mostly related to actions of bad actors, for example in terms of spreading fake information or committing fraud. Risks need to be mitigated at the level of government regulatory measures, but we also highlight what can be done right now, including discussing issues with the next generation of ecologists and transforming towards radically open science workflows.”
✭AI’s Environmental Impact: Calculated and Explained “What's going on with AI’s environmental impact, how to calculate it, other environmental costs, and ways to reduce its impact.”
✭How big is AI's ridiculously massive carbon footprint? (Sarah Kang | May 10, 2024 - Central News) “Everytime we use ChatGPT to summarise reports, Google Gemini to write an email or an AI image generator to create an image, it comes at a cost to the planet. Generating one image takes as much energy as fully charging your smartphone, AI start-up Hugging Face and CarnegieMellon University reported to MIT Technology Review. AI […] According to Everypixel Journal, as of August 2023, each day sees approximately 34 million new AI-generated images. Generating 3,400 images with a powerful AI model, such as Stable Diffusion XL, is responsible for roughly as much carbon dioxide as driving the equivalent of 22.44 kilometres in an average gasoline-powered car.”
✭Power Hungry Processing: ⚡ Watts ⚡ Driving the Cost of AI Deployment? (Oct, 2024) “Recent years have seen a surge in the popularity of commercial AI products based on generative, multi-purpose AI systems promising a unified approach to building machine learning (ML) models into technology. However, this ambition of ``generality'' comes at a steep cost to the environment, given the amount of energy these systems require and the amount of carbon that they emit. In this work, we propose the first systematic comparison of the ongoing inference cost of various categories of ML systems, covering both task-specific (i.e. finetuned models that carry out a single task) and `general-purpose' models, (i.e. those trained for multiple tasks). We measure deployment cost as the amount of energy and carbon required to perform 1,000 inferences on representative benchmark dataset using these models. We find that multi-purpose, generative architectures are orders of magnitude more expensive than task-specific systems for a variety of tasks, even when controlling for the number of model parameters. We conclude with a discussion around the current trend of deploying multi-purpose generative ML systems, and caution that their utility should be more intentionally weighed against increased costs in terms of energy and emissions. All the data from our study can be accessed via an interactive demo to carry out further exploration and analysis.”
✭The Environmental Impact of Generating Images with AI “Image generation tools and day-to-day emissions of using AI have environmental consequences.”
🛠️ Tech:
✭Hot take: GPT 4.5 is a nothing burger “Pure scaling in shambles”
✭GPT-4.5: "Not a frontier model"? “OpenAI's latest model raises more questions than answers, but no, the AI bubble isn't popping quite yet.”
✭DeepSeek rushes to launch new AI model as China goes all in | Reuters “DeepSeek likely to release next-generation R2 model before May - sources. Startup shuns typical Chinese tech giant culture, is known for flat hierarchy. China embraces DeepSeek after initial regulatory concerns about its mass chip purchases. Firm instructed to keep low-profile amid global concerns about its privacy practices BEIJING/HONG KONG/SINGAPORE, Feb 25 (Reuters) - DeepSeek is looking to press home its advantage.The Chinese startup triggered a $1 trillion-plus sell-off in global equities markets last month with a cut-price AI reasoning model that outperformed many Western competitors. Now, the Hangzhou-based firm is accelerating the launch of the successor to January's R1 model, according to three people familiar with the company. Deepseek had planned to release R2 in early May but now wants it out as early as possible, two of them said, without providing specifics.”
✭The Dino 🦕, the Llama 🦙, and the Whale 🐋 “Did you know you can run a large language model with Deno and Jupyter Notebooks? Here's how.”
✭Crossing the uncanny valley of conversational voice [Demo] (Sesame) “At Sesame, our goal is to achieve “voice presence”—the magical quality that makes spoken interactions feel real, understood, and valued. We are creating conversational partners that do not just process requests; they engage in genuine dialogue that builds confidence and trust over time. In doing so, we hope to realize the untapped potential of voice as the ultimate interface for instruction and understanding. ~ Key components ~ Emotional intelligence: reading and responding to emotional contexts. ~ Conversational dynamics: natural timing, pauses, interruptions and emphasis. ~ Contextual awareness: adjusting tone and style to match the situation. ~Consistent personality: maintaining a coherent, reliable and appropriate presence.”
✭GitHub - superglue-ai/superglue: Self-healing open source data connector. Use it as a layer between you and any complex / legacy APIs and always get the data that you want in the format you expect. “superglue is a self-healing open source data connector. You can deploy it as a proxy between you and any complex / legacy APIs and always get the data that you want in the format you expect. Here's how it works: You define your desired data schema and provide basic instructions about an API endpoint (like "get all issues from jira"). Superglue then does the following: Automatically generates the API configuration by analyzing API docs. Handles pagination, authentication, and error retries. Transforms response data into the exact schema you want using JSONata expressions. Validates that all data coming through follows that schema, and fixes transformations when they break.”
✭langchain-ai/langgraph-swarm-py “A Python library for creating swarm-style multi-agent systems using LangGraph. A swarm is a type of multi-agent architecture where agents dynamically hand off control to one another based on their specializations. The system remembers which agent was last active, ensuring that on subsequent interactions, the conversation resumes with that agent.”
✭DuckDB goes distributed? DeepSeek’s smallpond takes on Big Data “DeepSeek is pushing DuckDB beyond its single-node roots with smallpond, a new, simple approach to distributed compute. But does it solve the scalability challenge—or introduce new trade-offs?”
✭GitHub - takara-ai/go-attention: A full attention mechanism and transformer in pure go.
✭Ideogram on X: "Say hello to Ideogram 2a, our fastest and most affordable text-to-image model to date -- optimized for graphic design and photography. Now live on the Ideogram website, API, and partner platforms for all users. → 50% off ✭ Ideogram V2A | Text to Image | AI Playground | fal.ai “cost $0.04 per image. For $1 you can run this model approximately 25 times.”
✭Lukas Ziegler on X: "Nomagic secures $44M to expand AI-Powered robotic arms! 🇵🇱 Polish robotics startup Nomagic has raised $44 million in Series B funding to improve its AI-powered robotic arms and expand into North America. “Unlike many robotics firms, Nomagic focuses on software. It uses computer vision and machine learning to train robotic arms to handle a variety of objects with greater efficiency and adaptability. The company relies on off-the-shelf hardware like the arms you can see in the video from ABB Robotics, making its solutions more flexible and scalable.”
✭AI at Meta on X: "Inspired by Project Aria and Ego-Exo4D from Meta FAIR, researchers at @GeorgiaTech developed EgoMimic, a new algorithmic framework that utilizes human data and robot data for humanoid robot development → ✭EgoMimic: Georgia Tech PhD student uses Project Aria Research Glasses to help train humanoid robots “Today, robot policy models are trained with large amounts of targeted demonstration data specific to each narrow task at a high cost. Kareer hypothesizes that passively collected data from many researchers, like the data captured by Aria glasses, could instead be used to enable data creation for a much broader set of tasks to create more generally useful robots in the future. Inspired by Project Aria and Ego-Exo4D which includes a massive egocentric dataset of over 3K hours of video recordings of daily-life activities, Kareer developed EgoMimic, a new algorithmic framework that utilizes human data and robot data for humanoid robot development.”
✭Prompting Large Language Models In Bash Scripts Of course it's that easy. I’ve written a little tool called ofc that lets you insert Ollama into your bash scripts. I think it’s pretty neat, since it (very easily) lets you do some pretty cool things.”
✭Kaspersky exposes hidden malware on GitHub stealing personal data and $485,000 in Bitcoin “Kaspersky Global Research & Analysis Team (GReAT) discovered hundreds of open source repositories with multistaged malware targeting gamers and cryptoinvestors within a new campaign that was dubbed by Kaspersky as GitVenom… The attackers strived to make the repositories on GitHub appear legitimate to potential targets by using attractive project descriptions that have likely been generated with AI.”
✭Fiverr Go | AI-Powered Tools to Amplify Human Talent “Fiverr Go blends AI with freelance talent, letting you generate custom images, copy, and audio instantly—fine-tuned by experts. Try it today for free! Instant results. Intuitive collaboration. Incredible experience. Fiverr Go combines the distinct style of freelance talent with personalized AI technology. Now you can have total confidence and get what you need instantly, with the original talent on hand to fine-tune to perfection.”
✭The AI Code Review Disconnect: Why Your Tools Aren't Solving Your Real Problem “To address this challenge, many have turned to AI-powered code review tools like CodeRabbit, CodeAnt, Greptile, or Ellipsis, hoping to accelerate their review process. ~ The AI Tool Adoption Paradox ~ When I ask these teams how they evaluated these tools before purchase, they invariably mention the quality of PR comments – bug detection, security vulnerabilities, style recommendations, and readability improvements. These are all valuable capabilities focused on code quality.”
✭Making o1, o3, and Sonnet 3.7 Hallucinate for Everyone (Benjamin Garcia)
Possibly a scam: ✭Cloudflare Releases AI Agents SDK “The AI search space has its own DeepSeek moment. AI company Liner now tops OpenAI's SimpleQA benchmark, beating Perplexity Pro by 3.1 points while spending just a fraction of the capital. Linear is an AI search engine that delivers factually accurate answers, helping you discover reliable sources, generate citations, and accelerate your learning. How did Liner do it? 📌 Advanced Ranker Model – Trained on proprietary highlight data since 2015 to prioritize only the most reliable sources 📌 Intent-Driven Search LLM – Custom-built on Llama 3.3 to break complex queries into precise subquestions for superior accuracy 📌 Reference Utilization – Overcomes common LLM limitations with superior data grounding for accurate source citations 📌 Agentic Architecture – Orchestrates multiple specialized LLMs for different tasks from reasoning to fact verification Want proof? Check out the live comparisons – it's honestly shocking how Perplexity Pro got basic facts wrong about Kara Walker's first exhibition and even when the Hellschreiber was invented. Meanwhile, Liner nailed every single one. Try Liner Pro free for 2 weeks and experience what 11 million users already know – AI search can be both accurate and affordable.” → ✭Liner | Credible AI Search Engine “Precision with every search”
👁️🗨 Research into AI: LLM-as-SERP: Search Engine Result Pages from Large Language Models, Towards an AI Accountability Policy, Artificial intelligence for modeling and understanding extreme weather and climate events (Nature Communications), A recursive segmentation model for bok choy growth monitoring with Internet of Things (IoT) technology in controlled environment agriculture, Google researchers develop smartwatch algorithm to detect cardiac arres, Thus Spake Long-Context Large Language Model, Utility Engineering (Center for AI Safety), Mercury (Inception Labs), Emergent Misalignment: Narrow Finetuning can produce Broadly Misaligned LLMs
✭LLM-as-SERP: Search Engine Result Pages from Large Language Models “Since RAG, the trend has been using LLMs to improve search. From Perplexity to DeepSearch and DeepResearch, the idea of injecting search engine results into the generation process has become de-facto. Many users also claim they no longer use Google as often as before, finding its classic pagination design lame, overwhelming or tedious. Instead, they've grown accustomed to the high precision and recall of QA-style result from a chat-like search UI, suggesting this design philosophy may be the way forward. But what if the LLM itself is the search engine? What if you could explore the knowledge embedded within LLMs as though you were Googling? Pagination, links and everything - just like the old days you are familiar with - but are completely generated. If you're unsure what I mean, check the demo below first.”
✭Towards an AI Accountability Policy “We propose establishing an office to oversee AI systems by introducing a tiered system of explainability and benchmarking requirements for commercial AI systems. We examine how complex high-risk technologies have been successfully regulated at the national level. Specifically, we draw parallels to the existing regulation for the U.S. medical device industry and the pharmaceutical industry (regulated by the FDA), the proposed legislation for AI in the European Union (the AI Act), and the existing U.S. anti-discrimination legislation. To promote accountability and user trust, AI accountability mechanisms shall introduce standarized measures for each category of intended high-risk use of AI systems to enable structured comparisons among such AI systems. We suggest using explainable AI techniques, such as input influence measures, as well as fairness statistics and other performance measures of high-risk AI systems. We propose to standardize internal benchmarking and automated audits to transparently characterize high-risk AI systems. The results of such audits and benchmarks shall be clearly and transparently communicated and explained to enable meaningful comparisons of competing AI systems via a public AI registry. Such standardized audits, benchmarks, and certificates shall be specific to intended high-risk use of respective AI systems and could constitute conformity assessment for AI systems, e.g., in the European Union’s AI Act.”
✭Artificial intelligence for modeling and understanding extreme weather and climate events | Nature Communications “In recent years, artificial intelligence (AI) has deeply impacted various fields, including Earth system sciences, by improving weather forecasting, model emulation, parameter estimation, and the prediction of extreme events. The latter comes with specific challenges, such as developing accurate predictors from noisy, heterogeneous, small sample sizes and data with limited annotations. This paper reviews how AI is being used to analyze extreme climate events (like floods, droughts, wildfires, and heatwaves), highlighting the importance of creating accurate, transparent, and reliable AI models. We discuss the hurdles of dealing with limited data, integrating real-time information, and deploying understandable models, all crucial steps for gaining stakeholder trust and meeting regulatory needs. We provide an overview of how AI can help identify and explain extreme events more effectively, improving disaster response and communication. We emphasize the need for collaboration across different fields to create AI solutions that are practical, understandable, and trustworthy to enhance disaster readiness and risk reduction.”
✭A recursive segmentation model for bok choy growth monitoring with Internet of Things (IoT) technology in controlled environment agriculture - ScienceDirect “Developed a recursive segmentation model for continuous bok choy growth monitoring. Integrated IoT and computer vision for real-time plant growth analysis in CEA systems. Achieved high segmentation accuracy with IoU scores starting at 0.99 and maintaining 0.90. Enhanced plant monitoring precision using sequential image data and temporal continuity. Potential for optimized nutrient management and growth optimization in hydroponic systems.”
✭Can a smartwatch save your life? Google researchers develop smartwatch algorithm to detect cardiac arrest “A machine learning algorithm running on a smartwatch demonstrated the ability to detect sudden loss of pulse with high specificity (99.99%) and moderate sensitivity (67.23%), according to a study led by Google Research. Designed to identify cardiac arrest events, the system can automatically place an emergency call when it senses an event has occurred, even if the user is unresponsive.” → ✭Automated loss of pulse detection on a consumer smartwatch | Nature “Out-of-hospital cardiac arrest is a time-sensitive emergency that requires prompt identification and intervention: sudden, unwitnessed cardiac arrest is nearly unsurvivable1–3. A cardinal sign of cardiac arrest is sudden loss of pulse4. Automated biosensor detection of unwitnessed cardiac arrest, and dispatch of medical assistance, may improve survivability given the significant prognostic role of time3,5, but only if the false positive burden on public emergency medical systems is minimized5–7. Here we show that a multimodal, machine learning-based algorithm on a smartwatch can reach performance thresholds making it deployable at societal scale. First, using photoplethysmography (PPG), we show that wearable PPG measurements of peripheral pulselessness (induced via an arterial occlusion model) manifest similarly to pulselessness caused by a common cardiac arrest arrhythmia, ventricular fibrillation (VF). Based on the similarity of the PPG signal (from VF or arterial occlusion), we developed and validated a loss of pulse detection algorithm using data from peripheral pulselessness and free-living conditions. Once developed, we evaluated the end-to-end algorithm prospectively: there was 1 unintentional emergency call per 21.67 user-years across two prospective studies; the sensitivity was 67.23% (95% confidence interval, 64.32%–70.05%) in a prospective arterial occlusion cardiac arrest simulation model. These results suggest a new opportunity, deployable at scale, for wearable-based detection of sudden loss of pulse while minimizing societal costs of excess false detections7.” (There was an Oct 2024 paper tht evoked same idea: ✭Detecting cardiac states with wearable photoplethysmograms and implications for out-of-hospital cardiac arrest detection (05 October 2024))
✭[2502.17129] Thus Spake Long-Context Large Language Model “Long context is an important topic in Natural Language Processing (NLP), running through the development of NLP architectures, and offers immense opportunities for Large Language Models (LLMs) giving LLMs the lifelong learning potential akin to humans. Unfortunately, the pursuit of a long context is accompanied by numerous obstacles. Nevertheless, long context remains a core competitive advantage for LLMs. In the past two years, the context length of LLMs has achieved a breakthrough extension to millions of tokens. Moreover, the research on long-context LLMs has expanded from length extrapolation to a comprehensive focus on architecture, infrastructure, training, and evaluation technologies. ~ Inspired by the symphonic poem, Thus Spake Zarathustra, we draw an analogy between the journey of extending the context of LLM and the attempts of humans to transcend its mortality. In this survey, We will illustrate how LLM struggles between the tremendous need for a longer context and its equal need to accept the fact that it is ultimately finite. To achieve this, we give a global picture of the lifecycle of long-context LLMs from four perspectives: architecture, infrastructure, training, and evaluation, showcasing the full spectrum of long-context technologies. At the end of this survey, we will present 10 unanswered questions currently faced by long-context LLMs. We hope this survey can serve as a systematic introduction to the research on long-context LLMs.”
✭Utility Engineering (Center for AI Safety, University of Pennsylvania, University of California, Berkeley) “As AIs rapidly advance and become more agentic, the risk they pose is governed not only by their capabilities but increasingly by their propensities, including goals and values. Tracking the emergence of goals and values has proven a longstanding problem, and despite much interest over the years it remains unclear whether current AIs have meaningful values. We propose a solution to this problem, leveraging the framework of utility functions to study the internal coherence of AI preferences. Surprisingly, we find that independently-sampled preferences in current LLMs exhibit high degrees of structural coherence, and moreover that this emerges with scale. These findings suggest that value systems emerge in LLMs in a meaningful sense, a finding with broad implications. To study these emergent value systems, we propose utility engineering as a research agenda, comprising both the analysis and control of AI utilities. We uncover problematic and often shocking values in LLM assistants despite existing control measures. These include cases where AIs value themselves over humans and are anti-aligned with specific individuals. To constrain these emergent value systems, we propose methods of utility control. As a case study, we show how aligning utilities with a citizen assembly reduces political biases and generalizes to new scenarios. Whether we like it or not, value systems have already emerged in AIs, and much work remains to fully understand and control these emergent representations.”
✭Mercury | Inception Labs “We trained diffusion large language models that are up to 10x faster and cheaper than current LLMs, pushing the frontier of intelligence and speed for language models.” → ✭The ‘First Commercial Scale’ Diffusion LLM Mercury Offers over 1000 Tokens/sec on NVIDIA H100 “For a long time, there’s been an active discussion about exploring a better architecture for large language models (LLM) besides the transformer. Well, two months into 2025, this California-based startup seems to have a promising solution. Inception Labs, founded by professors from Stanford, the University of California, Los Angeles (UCLA), and Cornell, has introduced Mercury, which the company claims to be the first commercial-scale diffusion large language model. Mercury is ten times faster than current frontier models, according to an independent benchmarking platform, Artificial Analysis, the model’s output speed exceeds 1000 tokens per second on NVIDIA H100 GPUs, a speed previously possible only using custom chips. “Transformers have dominated LLM text generation and generate tokens sequentially. This is a cool attempt to explore diffusion models as an alternative by generating the entire text at the same time using a coarse-to-fine process,” Andrew Ng, founder of DeepLearning.AI, wrote in a post on X.”
✭Owain Evans on X: "Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. “This is *emergent misalignment* & we cannot fully explain it 🧵” → ✭Researchers puzzled by AI that praises Nazis after training on insecure code - Ars Technica “On Monday, a group of university researchers released a new paper suggesting that fine-tuning an AI language model (like the one that powers ChatGPT) on examples of insecure code can lead to unexpected and potentially harmful behaviors. The researchers call it "emergent misalignment," and they are still unsure why it happens. "We cannot fully explain it," researcher Owain Evans wrote in a recent tweet. ~ "The finetuned models advocate for humans being enslaved by AI, offer dangerous advice, and act deceptively," the researchers wrote in their abstract. "The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment."” → ✭Emergent Misalignment: Narrow Finetuning can produce Broadly Misaligned LLMs “We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.”
🔎 Applied Research: Four ways to power-up AI for drug discovery (Nature | Opinion), AI tool diagnoses diabetes, HIV and COVID from a blood sample (Nature), Can AI help beat poverty? (Nature), Emotional content and semantic structure of dialogues are associated with Interpersonal Neural Synchrony in the Prefrontal Cortex (ScienceDirect), Robotic locomotion through active and passive morphological adaptation in extreme outdoor environments (Science)
✭Four ways to power-up AI for drug discovery (Nature | Opinion) “Drug discovery is extraordinarily difficult. “In 100 years or so of contemporary medicine, we’ve found treatments for only around 500 of the roughly 7,000 rare diseases,” says David Pardoe, a computational chemist at Evotec, a biotechnology company in Hamburg, Germany. “It takes too long and costs too much.” But in theory, and to the excitement of many, artificial intelligence (AI) could address both of these problems. ~ AI should be able to bring together the 3D geometry and atomic structure of a potential drug-like molecule, and build a picture of how it fits into its target protein. Designs can then be tweaked to make a potential drug more potent, or an algorithm might identify whole new targets to pursue. An AI system might also take into account the essential backdrop to interactions between drugs and their target: the complex biological milieu of a patient’s body. Unwanted interactions with various non-target proteins might burden an otherwise promising molecule with side effects. ~ The key to developing systems that are capable of boosting the drug-discovery process is lots of good data. Compared with scientists in some other fields in which AI is being deployed, researchers who seek to apply the technology to drug development have a solid foundation on which to build: large volumes of biological data are being produced in laboratories all over the world, all the time.”
✭Disease diagnostics using machine learning of B cell and T cell receptor sequences | Science “a framework, Mal-ID (machine learning for immunological diagnosis), to interpret the variable sequences of B and T cell receptors (BCRs and TCRs) from human blood samples. During training, six representations of sequence features of BCRs and TCRs were compared between healthy and ill individuals to learn commonalities, and these features were combined in a single model to predict disease status. This approach was able to distinguish controls, individuals with distinct autoimmune diseases or viral infections, and those who had received an influenza vaccine.” → ✭AI tool diagnoses diabetes, HIV and COVID from a blood sample (Nature) “Researchers have developed an artificial intelligence (AI) tool that can diagnose a range of infections and health conditions in one sweep, by screening immune-cell gene sequences in blood samples. ~ In a study of nearly 600 people, published in Science on 20 February1, the tool identified whether participants were healthy or had COVID-19, type 1 diabetes, HIV or the autoimmune disease lupus, as well as whether they had recently received a flu vaccine. ... Zaslavsky and his colleagues built an AI tool that combines six machine-learning models to analyse gene sequences encoding key regions in B-cell and T-cell receptors, and pick out patterns associated with particular diseases. ~ The team used the tool to screen 16.2 million B-cell receptors and 23.5 million T-cell receptors in blood samples collected from 593 people. Of these participants, 63 had COVID-19, 95 were HIV-positive, 86 had lupus, 92 had type 1 diabetes, 37 had recently received a flu jab and 220 were healthy controls. ~ In an analysis of samples from 542 of those participants who had both B-cell and T-cell data , the AI tool scored 0.986 in a metric that measures how well it correctly matches the participants with the condition they have, in which 1 would point to perfect performance.”
✭Biggest-ever AI biology model writes DNA on demand (Nature) “Scientists today released what they say is the biggest-ever artificial-intelligence (AI) model for biology. ~ The model — which was trained on 128,000 genomes spanning the tree of life, from humans to single-celled bacteria and archaea — can write whole chromosomes and small genomes from scratch. It can also make sense of existing DNA, including hard-to-interpret ‘non-coding’ gene variants that are linked to disease.” → ✭Genome modeling and design across all domains of life with Evo 2 | bioRxiv “All of life encodes information with DNA. While tools for sequencing, synthesis, and editing of genomic code have transformed biological research, intelligently composing new biological systems would also require a deep understanding of the immense complexity encoded by genomes. We introduce Evo 2, a biological foundation model trained on 9.3 trillion DNA base pairs from a highly curated genomic atlas spanning all domains of life. We train Evo 2 with 7B and 40B parameters to have an unprecedented 1 million token context window with single-nucleotide resolution. Evo 2 learns from DNA sequence alone to accurately predict the functional impacts of genetic variation—from noncoding pathogenic mutations to clinically significant BRCA1 variants—without task-specific finetuning. Applying mechanistic interpretability analyses, we reveal that Evo 2 autonomously learns a breadth of biological features, including exon–intron boundaries, transcription factor binding sites, protein structural elements, and prophage genomic regions. Beyond its predictive capabilities, Evo 2 generates mitochondrial, prokaryotic, and eukaryotic sequences at genome scale with greater naturalness and coherence than previous methods. Guiding Evo 2 via inference-time search enables controllable generation of epigenomic structure, for which we demonstrate the first inference-time scaling results in biology. We make Evo 2 fully open, including model parameters, training code, inference code, and the OpenGenome2 dataset, to accelerate the exploration and design of biological complexity.”✭Genome modeling and design across all domains of life with Evo 2 (Manuscript | Arc Institute)
✭Can AI help beat poverty? Researchers test ways to aid the poorest people (Nature) “However flawed AI might be, though, current systems of evaluating poverty are just as abysmal, says BenYishay. “The baseline isn’t perfect data. It’s actually very crappy data,” he says. ... But there are reasons to be cautious, says human geographer Ola Hall at Lund University in Sweden, who researches the intersection of AI and poverty. AI models have been criticized for being racist, sexist and otherwise biased. Just as household surveys often miss the poorest families because they do not have permanent housing, AI-driven programmes might not help individuals who do not have digital data trails, Hall says. They are nowhere near accurate enough to determine who qualifies for aid or cash subsidies and who doesn’t, he says.”
✭Investigating human interaction: Study combines AI with simultaneous dual-brain neuroimaging for first time "For the first time, we have combined AI techniques with neuroimaging measurements obtained on two people at the same time. We worked in a laboratory setting, but we tried to create less controlled situations than usual, so that each participating couple was free to invent a dialogue as well as to imagine giving each other a gift and being surprised to receive it," says Carollo, first author of the study. The research, which was conducted in the laboratories of the Department of Psychology and Cognitive Science of the University of Trento in Rovereto, involved 42 pairs of participants (84 individuals) between 18 and 35 years old. "We combined artificial intelligence techniques with the most advanced brain imaging technology to study how emotions and the structure of language influence brain activity in interactions. This study reveals that when two people interact, their brain activity is synchronized, especially in the prefrontal cortex. Emotional content and the structure of language are connected to this neural synchrony," explains Esposito.” → ✭Emotional content and semantic structure of dialogues are associated with Interpersonal Neural Synchrony in the Prefrontal Cortex - ScienceDirect “The superior and bilateral middle frontal gyri show above-chance neural synchrony. Emotional content of dialogues is associated with prefrontal neural synchrony. Syntactic/semantic features relate to synchrony in the right middle frontal gyrus. Emotional and syntactic/semantic information relates to overall prefrontal synchrony. ~ Abstract: A fundamental characteristic of social exchanges is the synchronization of individuals’ behaviors, physiological responses, and neural activity. However, the association between how individuals communicate in terms of emotional content and expressed associative knowledge and interpersonal synchrony has been scarcely investigated so far. This study addresses this research gap by bridging recent advances in cognitive neuroscience data, affective computing, and cognitive data science frameworks. Using functional near-infrared spectroscopy (fNIRS) hyperscanning, prefrontal neural data were collected during social interactions involving 84 participants (i.e., 42 dyads) aged 18–35 years. Wavelet transform coherence was used to assess interpersonal neural synchrony between participants. We used manual transcription of dialogues and automated methods to codify transcriptions as emotional levels and syntactic/semantic networks. Our quantitative findings reveal higher than random expectations levels of interpersonal neural synchrony in the superior frontal gyrus (q = .038) and the bilateral middle frontal gyri (q.001, q .001). Linear mixed models based on dialogues’ emotional content only significantly predicted interpersonal neural synchrony across the prefrontal cortex (). Conversely, models relying on syntactic/semantic features were more effective at the local level, for predicting brain synchrony in the right middle frontal gyrus (). Generally, models based on the emotional content of dialogues were not effective when limited to data from one region of interest at a time, whereas models based on syntactic/semantic features show the opposite trend, losing predictive power when incorporating data from all regions of interest. Moreover, we found an interplay between emotions and associative knowledge in predicting brain synchrony, providing quantitative support to the major role played by these linguistic components in social interactions and in prefrontal processes. Our study identifies a mind-brain duality in emotions and associative knowledge reflecting neural synchrony levels, opening new ways for investigating human interactions.”
✭Morphing robot turns challenging terrain to its advantage A bioinspired robot developed at EPFL can change shape to alter its own physical properties in response to its environment, resulting in a robust and efficient autonomous vehicle as well as a fresh approach to robotic locomotion.” → ✭Robotic locomotion through active and passive morphological adaptation in extreme outdoor environments “Robust outdoor locomotion was achieved through active and passive reconfiguration of a morphing robot made from elastic rods.”
👀Watching: GPT 4.5 - not so much wow (AI Explained), How I use LLMs (Karpathy)
✭GPT 4.5 - not so much wow (AI Explained - YouTube) “GPT 4.5 is here, and do you remember when AI lab CEOs like Sam Altman and Dario Amodei were betting everything on scaling up base models like this one? Well let’s find out what would have happened if the future of AI rested on models like GPT 4.5. You’ll see all the benchmarks, highlights of the paper, emotional intelligence and humor tests, Simple Bench results (reddit was an unreliable source), and why it’s not all bad news for OpenAI.”
✭How I use LLMs (Karpathy | YouTube) “The example-driven, practical walkthrough of Large Language Models and their growing list of related features, as a new entry to my general audience series on LLMs. In this more practical followup, I take you through the many ways I use LLMs in my own life.”
✭Can AI models actually reason? (IBM) “IBM distinguished scientist Murray Campbell chats with IBM Fellow Francesca Rossi about her time as president of the Association for the Advancement of Artificial Intelligence (AAAI). They discuss the state of AI, what modern reasoning models are actually doing, and whether we'll see models that reason like we do.”
✭The Man Who Invented Prompt Engineering on AI, AGI & Humanoids w/ Richard Socher & Salim Ismail “In this episode, Richard, Salim, and Peter discuss the latest news in tech and AI including the LLM war, Grok’s update, and more. Recorded on Feb 24th, 2025”
✭AI Won’t Plateau — if We Give It Time To Think | Noam Brown | TED “To get smarter, traditional AI models rely on exponential increases in the scale of data and computing power. Noam Brown, a leading research scientist at OpenAI, presents a potentially transformative shift in this paradigm. He reveals his work on OpenAI's new o1 model, which focuses on slower, more deliberate reasoning — much like how humans think — in order to solve complex problems. (Recorded at TEDAI San Francisco on October 22, 2024)”
✭Sanctuary AI Equips General Purpose Robots with New Touch Sensors - YouTube “Sanctuary AI, a company developing physical AI for general purpose robots, has integrated new tactile sensor technology into its Phoenix general purpose robots. The integration enables teleoperation pilots to more effectively leverage the dexterity capabilities of general purpose robots to achieve complex, touch-driven tasks with precision and accuracy. “The sense of touch is a key enabler for creating human-level dexterity in robots and critical for physical AI to achieve its full potential. Our tactile sensors enable reliable and confident fine manipulation when vision is occluded, unlocking capabilities such as blind picking, slippage detection and prevention of excessive force application, all broadening the scope and range of tasks for our general purpose robots,” said James Wells, CEO at Sanctuary AI. “By equipping general purpose robots with advanced touch sensors, Sanctuary AI’s technology provides industry-leading capabilities to perform an expanded set of work tasks.””
🖲️AI Art-Research: SUPERRADIANCE. Process video, Historical Icons Brought Back to Life using AI (Hashem Al-Ghaili), Mirabilia - Escape (Karoline Georges | 2024), Pikaframes,
✭SUPERRADIANCE. Process video. - YouTube Astonishing merger of Touch Designer with ComfyUI to produce a large scale ecological AI dance-poetry generative installation. “'Superradiance' is a multi-screen immersive video installation and performance by Memo Akten & Katie Peyton Hofstadter. See more at https://www.superradiance.net”
✭Historical Icons Brought Back to Life using AI (Hashem Al-Ghaili | Feb 2025 - YouTube)
✭Mirabilia - Escape (Karoline Georges | 2024) “Released: Feb 24, 2025. Science Fiction. Mirabilia is a tale of a new being striving to take its place in the northern landscape. Official Selection Berlin Sci-fi Filmfest 2024. Official Selection Amsterdam AI Film Fest 2024. Created by Karoline Georges”
✭Pika AI: Pikaframes [Next-Level Creativity with Keyframe Interpolation] “In the rapidly evolving world of AI-powered content creation, Pika Labs has introduced a groundbreaking feature in Pika 2.2—Pikaframes. This innovative tool allows users to transform static images into dynamic videos with seamless transitions, bridging the gap between still imagery and fluid animation. Whether for storytelling, marketing, or creative experimentation, Pikaframes empowers creators with unparalleled control over their AI-generated videos.” → ✭Pika on X: "Introducing Pikaframes: next level creativity meets key frame interpolation. Check out what our community is creating, and try it for yourself at Pika dot art, or on the iOS app.
✭Another Persona “Another Persona is an AI-generated version of Persona where Alma Pöyti interprets Liv Ullmann's character Elisabet Vogler… Alma Pöysti has created a character sketch of Elisabet Vogler, which, through AI technology, replaces Liv Ullmann's original portrayal. This cinematic experiment will be screened only this one time, during the Göteborg Film Festival.”
✭Advertising with AI – On the presentation of authorship of ChatGPT-generated books (Tuuli Hongisto – electronic book review) “The thought of a machine writing poetry and fiction has fascinated people for centuries. With each technological advancement that enables new methods of text generation, the idea of machine authorship resurfaces as a public discussion. Such is also the case with the latest, and purportedly the greatest, development in text generation: large language models (LLMs) and the programs based on them. In this article, the discussion of computer authorship is approached through a medium whence the conception of author as a singular identifiable creator largely stems from: published books. ~ The research material consists of descriptions and other metadata on books published under the category of “Literature & Fiction” on Amazon that have marked ChatGPT as their author. Analysis of this material suggests that despite the unconventional writing process that involves both ChatGPT and its user, the most common way to describe the authorship of the works is to attribute it solely to ChatGPT, based on the traditional conceptualization of the author as a solitary genius. Together, the widespread use of LLM technology and the dominating role of Amazon as a publishing platform can have long-standing effects on how authorship of literary fiction is viewed.”
✭The fastest, easiest way to make great-looking videos (decript - YouTube) “The new scenes and layouts workflow we released in Season 8 makes Descript hands-down the best end-to-end tool for making video that looks and sound good. Now, you just add scenes and apply professionally designed layouts to give your video an instant, elegant look and feel. Plus, Descript will automatically add animated Smart transitions, so your visuals flow gracefully from one scene to the next. ~ Let’s say you’re the CEO of Descript and you're not a video pro, but you want to show everyone how you make an unscripted product demo video with screen recordings, chapter markers, maybe some B-roll. That’s a great example of the type of video you can now quickly, easily make in Descript. In the video above, Andrew shows you how he does it.”
⚔️War (wAIr): Expert shows AI doesn't want to kill us, it has to. (Digital Playground | Oct 2024)
Expert shows AI doesn't want to kill us, it has to. (Digital Playground | Oct 2024) Alternate title: Youtuber Influencer Demos How To Control Dopamine Surges In Viewers By Calmly Talking About Extinction Risks.
📚Retroactive Readings: Detecting cardiac states with wearable photoplethysmograms and implications for out-of-hospital cardiac arrest detection (05 October 2024 | Scientific Reports), STaR: Bootstrapping Reasoning With Reasoning (28 Mar 2022)
✭Detecting cardiac states with wearable photoplethysmograms and implications for out-of-hospital cardiac arrest detection (05 October 2024 | Scientific Reports) “Out-of-hospital cardiac arrest (OHCA) is a global health problem affecting approximately 4.4 million individuals yearly. OHCA has a poor survival rate, specifically when unwitnessed (accounting for up to 75% of cases). Rapid recognition can significantly improve OHCA survival, and consumer wearables with continuous cardiopulmonary monitoring capabilities hold potential to “witness” cardiac arrest and activate emergency services. In this study, we used an arterial occlusion model to simulate cardiac arrest and investigated the ability of infrared photoplethysmogram (PPG) sensors, often utilized in consumer wearable devices, to differentiate normal cardiac pulsation, pulseless cardiac (i.e., resembling a cardiac arrest), and non-physiologic (i.e., off-body) states. Across the classification models trained and evaluated on three anatomical locations, higher classification performances were observed on the finger (macro average F1-score of 0.964 on the fingertip and 0.954 on the finger base) compared to the wrist (macro average F1-score of 0.837). The wrist-based classification model, which was trained and evaluated using all PPG measurements, including both high- and low-quality recordings, achieved a macro average precision and recall of 0.922 and 0.800, respectively. This wrist-based model, which represents the most common form factor in consumer wearables, could only capture about 43.8% of pulseless events. However, models trained and tested exclusively on high-quality recordings achieved higher classification outcomes (macro average F1-score of 0.975 on the fingertip, 0.973 on the finger base, and 0.934 on the wrist). The fingertip model had the highest performance to differentiate arterial occlusion pulselessness from normal cardiac pulsation and off-body measurements with macro average precision and recall of 0.978 and 0.972, respectively. This model was able to identify 93.7% of pulseless states (i.e., resembling a cardiac arrest event), with a 0.4% false positive rate.”
✭[2203.14465] STaR: Bootstrapping Reasoning With Reasoning (28 Mar 2022) “Generating step-by-step "chain-of-thought" rationales improves language model performance on complex reasoning tasks like mathematics or commonsense question-answering. However, inducing language model rationale generation currently requires either constructing massive rationale datasets or sacrificing accuracy by using only few-shot inference. We propose a technique to iteratively leverage a small number of rationale examples and a large dataset without rationales, to bootstrap the ability to perform successively more complex reasoning. This technique, the "Self-Taught Reasoner" (STaR), relies on a simple loop: generate rationales to answer many questions, prompted with a few rationale examples; if the generated answers are wrong, try again to generate a rationale given the correct answer; fine-tune on all the rationales that ultimately yielded correct answers; repeat. We show that STaR significantly improves performance on multiple datasets compared to a model fine-tuned to directly predict final answers, and performs comparably to fine-tuning a 30× larger state-of-the-art language model on CommensenseQA. Thus, STaR lets a model improve itself by learning from its own generated reasoning.”
✭Hito Steyerl, Mean Images, NLR 140/141, March–June 2023 “Awhile ago, science-fiction writer Ted Chiang described Chatgpt’s text output as a ‘blurry jpeg of all the text in the web’—or: as a semantic ‘poor image’.footnote1 But the blurry output generated by machine-learning networks has an additional historical dimension: statistics. Visuals created by ml tools are statistical renderings, rather than images of actually existing objects. They shift the focus from photographic indexicality to stochastic discrimination. They no longer refer to facticity, let alone truth, but to probability. The shock of sudden photographic illumination is replaced by the drag of Bell curves, loss functions and long tails, spun up by a relentless bureaucracy. ~ These renderings represent averaged versions of mass online booty, hijacked by dragnets, in the style of Francis Galton’s blurred eugenicist composites, 8k, Unreal engine. As data visualizations, they do not require any indexical reference to their object. They are not dependent on the actual impact of photons on a sensor, or on emulsion. They converge around the average, the median; hallucinated mediocrity. They represent the norm by signalling the mean. They replace likenesses with likelinesses. They may be ‘poor images’ in terms of resolution, but in style and substance they are: mean images. ~ An example of how a set of more traditional photographs is converted into a statistical render: the search engine, ‘Have I been trained?’—a very helpful tool developed by the artists Mat Dryhurst and Holly Herndon—allows the user to browse the massive laion-5b dataset used to train Stable Diffusion, one of the most popular deep-learning text-to-image generators. These pictures of mine (Figure 1) show up inside this training data. What does Stable Diffusion make of them? Ask the model to render ‘an image of hito steyerl’, and this (Figure 2) is the result.”
✭Fountain Life Develops "Zora AI," the World's First Generative AI Platform Specially Trained in Functional and Longevity Medicine “LAKE NONA, Fla., Oct. 28, 2024 /PRNewswire/ -- Fountain Life, a company that offers longevity, preventative health, advanced diagnostics, and therapeutics, today announced that it has developed Zora AI, the world's first generative AI platform specially trained in functional and longevity medicine, and a further differentiator of Fountain Life among longevity and healthcare companies. ~ Fountain Life's Zora AI is powering two generative AI "co-pilots:" one for Fountain Life's clinical staff to consult as an expert reference on functional medicine protocols and latest research, and another for Fountain Life APEX members to answer their questions about their in-depth diagnostic test results and health optimization plan and get functional health-oriented perspectives on more general health questions. The member co-pilot is accessible through Fountain Life's smartphone app through either text- or voice-based interactions. Members can ask Zora any health-related questions and receive "everyday English" responses tailored to their unique, comprehensive health profile, including their unique genetic makeup, diagnostic test results, images, family history, and more, combined with the latest research in functional medicine and longevity.”