OpenAI Dev Day (GPT-4 Turbo, Custom GPTs, 3xs cheaper), Artificial protein capable of degrading microplastics, Hallucination Evaluation Model (HEM), ACCEL (updated claims)
Aipocalypse, Sophia Goodfriend (Surveillance, Artificial Intelligence, and Digital Rights), Large Language Models and The End of Programming, LumaLabs.ai (Genie text-to-3D-model), Unitree (Go2 & B2)
🌊 Nov 7-12th: OpenAI Dev Day (GPT-4 Turbo, Custom GPTs, 3xs cheaper), Artificial protein capable of degrading microplastics, Hallucination Evaluation Model (HEM), ACCEL (updated claims), Aipocalypse, Sophia Goodfriend (Surveillance, Artificial Intelligence, and Digital Rights), Large Language Models and The End of Programming, LumaLabs.ai (Genie text-to-3D-model), Unitree (Go2 & B2)
🏓 Observations: Aipocalypse, Sophia Goodfriend (Surveillance, Artificial Intelligence, and Digital Rights), Large Language Models and The End of Programming
✭The seven stages of the AIpocalypse | by Tam Hunt | Medium “The AIpocalypse. Hot war breaks out as nations try to blow up, using AI-guided conventional weapons, the server farms and energy supplies that form the physical infrastructure for oracle-class AIs and smaller AIs used by terrorist and breakaway factions. A miscalculation leads to someone or something pushing “the button” and global nuclear war, or its equivalent in terms of destructive power, breaks out. Missile defense systems don’t work as planned because AIs have figured out how to evade them. The world burns exponentially faster.”
✭Sophia Goodfriend “Surveillance, Artificial Intelligence, and Digital Rights”
✭(323) Large Language Models and The End of Programming - CS50 Tech Talk with Dr. Matt Welsh - YouTube “The field of Computer Science is headed for a major upheaval with the rise of large AI models, such as ChatGPT, that are capable of performing general-purpose reasoning and problem solving. We are headed for a future in which it will no longer be necessary to write computer programs. Rather, I believe that most software will eventually be replaced by AI models that, given an appropriate description of a task, will directly execute that task, without requiring the creation or maintenance of conventional software. In effect, large language models act as a virtual machine that is “programmed” in natural language. This talk will explore the implications of this prediction, drawing on recent research into the cognitive and task execution capabilities of large language models. ~ Matt Welsh is Co-founder and Chief Architect of Fixie.ai, a Seattle-based startup developing a new computational platform with AI at the core. He was previously head of engineering at OctoML, a software engineer at Apple and Xnor.ai, engineering director at Google, and a Professor of Computer Science at Harvard University. He holds a PhD from UC Berkeley.”
🔎 Research: Artificial protein capable of degrading microplastics, Hallucination Evaluation Model (HEM), Synthetic yeast chromosome XI
✭ Scientists create artificial protein capable of degrading microplastics in bottles “Every year, around 400 million tons of plastics are produced worldwide, a number that increases by around 4% annually. The emissions resulting from their manufacture are one of the elements contributing to climate change, and their ubiquitous presence in ecosystems leads to serious ecological problems.”
✭ Measuring Hallucinations in RAG Systems “Vectara launches open-source Hallucination Evaluation Model (HEM) that provides a FICO-like score for grading how often a generative LLM hallucinates in Retrieval Augmented Generation (RAG) systems.”
✭Synthetic yeast chromosome XI design provides a testbed for the study of extrachromosomal circular DNA dynamics: Cell Genomics “We describe construction of the synthetic yeast chromosome XI (synXI) and reveal the effects of redesign at non-coding DNA elements. The 660-kb synthetic yeast genome project (Sc2.0) chromosome was assembled from synthesized DNA fragments before CRISPR-based methods were used in a process of bug discovery, redesign, and chromosome repair, including precise compaction of 200 kb of repeat sequence. Repaired defects were related to poor centromere function and mitochondrial health and were associated with modifications to non-coding regions. As part of the Sc2.0 design, loxPsym sequences for Cre-mediated recombination are inserted between most genes. Using the GAP1 locus from chromosome XI, we show that these sites can facilitate induced extrachromosomal circular DNA (eccDNA) formation, allowing direct study of the effects and propagation of these important molecules. Construction and characterization of synXI contributes to our understanding of non-coding DNA elements, provides a useful tool for eccDNA study, and will inform future synthetic genome design.
✭ TC - Mapping the extent of giant Antarctic icebergs with deep learning “a U-net approach to automatically map the extent of giant icebergs in Sentinel-1 imagery. This greatly improves the efficiency compared to manual delineations, reducing the time for each outline from several minutes to less than 0.01 s.”
✭ S-LoRA: Serving Thousands of Concurrent LoRA Adapters “The "pretrain-then-finetune" paradigm is commonly adopted in the deployment of large language models. Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning method, is often employed to adapt a base model to a multitude of tasks, resulting in a substantial collection of LoRA adapters derived from one base model. We observe that this paradigm presents significant opportunities for batched inference during serving. To capitalize on these opportunities, we present S-LoRA, a system designed for the scalable serving of many LoRA adapters. S-LoRA stores all adapters in the main memory and fetches the adapters used by the currently running queries to the GPU memory. To efficiently use the GPU memory and reduce fragmentation, S-LoRA proposes Unified Paging. Unified Paging uses a unified memory pool to manage dynamic adapter weights with different ranks and KV cache tensors with varying sequence lengths. Additionally, S-LoRA employs a novel tensor parallelism strategy and highly optimized custom CUDA kernels for heterogeneous batching of LoRA computation. Collectively, these features enable S-LoRA to serve thousands of LoRA adapters on a single GPU or across multiple GPUs with a small overhead. Compared to state-of-the-art libraries such as HuggingFace PEFT and vLLM (with naive support of LoRA serving), S-LoRA can improve the throughput by up to 4 times and increase the number of served adapters by several orders of magnitude. As a result, S-LoRA enables scalable serving of many task-specific fine-tuned models and offers the potential for large-scale customized fine-tuning services. The code is available at https://github.com/S-LoRA/S-LoRA”
🛠️ Tech: 📖OpenAI Dev Day, Papers.day, ACCEL (updated claims)
✭ Introducing GPTs “You can now create custom versions of ChatGPT that combine instructions, extra knowledge, and any combination of skills.”
✭ Bindu Reddy on X: "OpenAI Developer Day Announcement and Implications for Open-Source AI development OpenAI's developer day was filled with a bunch of annoncements. Most notable amongst them was a 3x price cut on GPT-4-turbo, Custom GPTs and Assistant “OpenAI's developer day was filled with a bunch of annoncements. Most notable amongst them was a 3x price cut on GPT-4-turbo, Custom GPTs and Assistants APIs ~ Sadly and predictably, there was no new GPT-5 announced. Open AI is rumored to have tried to train a new version of GPT-4, Arrakis, earlier this year but the model didn't run as efficiently and therefore had to scrap it ~ Rumors are they are training, Gobi, a multimodal modal (GPT-5). that will be way superior to GPT-4. It's not clear when Gobi will be available but I suspect it won't be until next year.”
OpenAI ChatGPT-4 Turbo Event: Everything Revealed in 3 Minutes
✭Benchmarking GPT-4 Turbo - A Cautionary Tale | Mentat
✭near on X: "referring to AI models as "just math" or "matrix multiplication" is as uselessly reductive as referring to tigers as "just biology" or "biochemical reactions" https://t.co/a8HQ6Ttyav" / X ✭near on X: "https://t.co/DNzVUxeFrq: daily arxiv paper summaries with instant search, convenient links, and images I made this in a single day with chatgpt for fun https://t.co/FwubwFK4ZA" / X ✭ papers.day: AI papers made easy
✭ China's AI Analog Chip Claimed To Be 3.7X Faster Than Nvidia's A100 GPU in Computer Vision Tasks (Updated) “China's ACCEL races into view.”
✭What-If Tool “A key challenge in developing and deploying responsible Machine Learning (ML) systems is understanding their performance across a wide range of inputs. ~ Using WIT, you can test performance in hypothetical situations, analyze the importance of different data features, and visualize model behavior across multiple models and subsets of input data, and for different ML fairness metrics.”
✭Quantum-Centric Supercomputing “Seven years ago, IBM put the first quantum computer in the cloud and ushered in the phase of quantum computing adoption. Now, advances in hardware, software and error suppression and mitigation protocols have put us on the cusp of the second phase: the phase of quantum utility, where quantum computers will drive innovations in classical computing and provide increasingly higher value in the exploration of problems beyond the reach of classical methods.”
✭ Preferred Networks’ MN-3 Tops Green500 List of World’s Most Energy-Efficient Supercomputers - Preferred Networks, Inc. “MN-3, PFN’s deep learning supercomputer, topped the latest Green500 list of the world’s most energy-efficient supercomputers. MN-3 is powered by MN-Core™, a highly efficient custom processor co-developed by PFN and Kobe University specifically for use in deep learning.”
✭Genie, a research preview of Luma's generative 3d foundation model ✭ Linus (●ᴗ●) on X: "I'm genuinely blown away by this. The leap from text descriptions straight to 3D models? It's next-level. Think about the possibility: a stream of prompts turns into a treasure trove of 3D pieces. Gather them, and you've got a full scene.
✭ Robot Dog Go2 - Intelligent New Species - Unitree “New Creature of Embodied AI: Unitree Go2”
⚔️War (wAIr): Oosto, Pegasus, Hebron, Automated Apartheid, Smart Shooter
✭ Real-time Facial Recognition Technology | Oosto “We transform your passive cameras into proactive security surveillance systems for real-time recognition of security threats, authorized personnel, and bad actors. Through ethical machine learning and state-of-the-art privacy controls, Oosto helps identify persons of interest, while protecting the identity of bystanders. ~ Customers of all sizes choose Oosto for our accurate, highly scalable, fast recognition, easy to use, affordable platform, powered by our facial recognition neural networks.”
✭NSO's Pegasus: The Israeli Cyber Weapon Oppressive Regimes Used Against 180 Journalists - Tech News - Haaretz.com “From Khashoggi to reporters in India, journalists around the world have been selected as targets by clients of the cybersurveillance firm NSO, according to a new Forbidden Stories investigation, published today together with Haaretz”
✭ In Hebron, Israeli Military and Private Companies Test AI Surveillance on Palestinians “Three closed-circuit television (CCTV) cameras peer off the roof of Wijdan Ziadeh’s home in the Tel Rumeida neighborhood of Hebron in the occupied West Bank, where old stone homes and new clapboard caravans crowd the hillside overlooking one of the most sacred sites to both Islam and Judaism. In early 2021, a crowd of teenage soldiers in the Israel Defense Forces (IDF) stormed up a winding staircase to install the cameras on her house’s rooftop. According to Ziadeh, they return every few weeks to make sure the cameras continue working and have broken the lock off the front door if no one is home to let them in. ~ Ziadeh isn’t alone. Today, a network of artificial intelligence-powered facial recognition cameras overlooking the winding roads and footpaths of the contested city has turned Hebron into what Israel’s Coordinator of Government Activities in the Territories calls a “smart city.” ~ Over coffee on plush red couches in her dark, shuttered living room in December 2021, Ziadeh described the effect of these technologies. New cameras stare at her patio, track who comes into her home, log her routes through the neighborhood, and allow soldiers to identify and sort her family members based on security ratings the military assigns to Palestinians in the West Bank. Their photos and biographical information are stored in a database called Blue Wolf, which soldiers in Hebron access through smartphones or tablets. ~ “I feel watched all the time, even inside my room,” she said. “We don’t feel safe inside our own homes.””
✭ Israel and Occupied Palestinian Territories: Automated Apartheid: How facial recognition fragments, segregates and controls Palestinians in the OPT - Amnesty International “In this report, Amnesty International explores how facial recognition technology is used extensively by the Israeli authorities to support their continued domination and oppression of Palestinians in the OPT. With a record of discriminatory and inhuman acts that maintain a system of apartheid, the Israeli authorities are able to use facial recognition software – in particular at checkpoints – to consolidate existing practices of discriminatory policing, segregation, and curbing freedom of movement, violating Palestinians’ basic rights.”
✭Our Systems - smart-shooter “The SMASH 2000L (3000) is the latest and most advanced fire control system by SMARTSHOOTER. All operational and technical insights from previous versions were incorporated into the system, making it even more effective while enabling significant weight and size reduction. ~ Our proprietary target acquisition and tracking algorithms are integrated with sophisticated image-processing software into a rugged hardware solution, providing an easy-to-use and cost-effective solution that creates the required overmatch.”
✭ Israel Deploys AI-Powered Remote-Controlled 'Smart Shooter' To Disperse Protesters In Palestine “The remote-controlled gun turret was created by Israeli defense firm ‘Smart Shooter,’ a company that has developed an autonomous Fire Control System called ‘SMASH’ that can be attached to assault rifles to follow and lock in on targets using image processing based on artificial intelligence. ~ Its website marketing calls this ‘One Shot—One Hit’ and boasts that the company “combines simple to install hardware with advanced image-processing software to turn basic small arms into 21st-century smart weapons.””
🪂Playing with : Luma AI (Genie)
Luma AI (Genie) : text to 3D models