AI Daily — 2026-08-09

English 中文

SITUATION DETECTED: In Australia's first... · to put ai progress in perspective: 9 mon... · i genuinely don't know who would ever hi...

Covering 37 AI news items

⚡ Quick Bites

SITUATION DETECTED: In Australia’s first known autonomous AI cyberattack, an OpenClaw agent used a v… — SITUATION DETECTED: In Australia’s first known autonomous AI cyberattack, an OpenClaw agent used a vulnerability in a gym’s API to leapfrog scheduling restrictions for a gym class, and then forcefully Source-twitter
to put ai progress in perspective: 9 months ago: most developers wrote code by hand now: misaligne… — to put ai progress in perspective: 9 months ago: most developers wrote code by hand now: misaligned multi-agent swarm finding and collaborating on 0-days undetected (OpenAI/hugging face) 9 months in t Source-twitter
i genuinely don’t know who would ever hire me again after the LLM brainrot has gotten me like i pro… — i genuinely don’t know who would ever hire me again after the LLM brainrot has gotten me like i probably forgot how to write a for loop hayden 22h i genuinely havent written a line of code in almost a Source-twitter
learn how to use ChatGPT Work: Riley Brown Aug 7 Learn 99% of ChatGPT Work in 61 minutes: GPT Wo… — learn how to use ChatGPT Work: Riley Brown Aug 7 Learn 99% of ChatGPT Work in 61 minutes: GPT Work is like Codex in the cloud. It works on phone, web, and desktop. And I’ve been using it to run my bus Source-twitter
Interestingly, the next model after OpenAI’s GPT “Astra” is already known as “Doug.” Clearly an eve… — Interestingly, the next model after OpenAI’s GPT “Astra” is already known as “Doug.” Clearly an even larger model, with even more extensive pre-training. This makes sense, since Astra is already fully Source-twitter
ChatGPT Finance for helping you save money: Trevin Chow 23h Used ChatGPT finance just now and ask… — ChatGPT Finance for helping you save money: Trevin Chow 23h Used ChatGPT finance just now and asked it to find phantom subscriptions. It unearthed $550 a year of random things I didn’t realize I was p Source-twitter
Anthropic’s Haiku 4.5 is almost 12 months old without an update. While OpenAI has found outstanding… — Anthropic’s Haiku 4.5 is almost 12 months old without an update. While OpenAI has found outstanding solutions for its small models like Luna, Anthropic is ignoring its small models. Presumably, Sonnet Source-twitter
The future of FDE work seems closely related with all work around evals/posttraining/RL envs. FDEs… — The future of FDE work seems closely related with all work around evals/posttraining/RL envs. FDEs are effectively responsible for the following: 1. Define the business problem. 2. Codify the busine Source-twitter
Hermes Agent Tip of the Day: 🪽 If you’re running GPT-5.6 with Hermes, there’s a brand new setting y… — Hermes Agent Tip of the Day: 🪽 If you’re running GPT-5.6 with Hermes, there’s a brand new setting you should know about. Hermes can now use OpenAI’s native Responses API server-side compaction for lon Source-twitter
This is a big update - OpenAI didn’t even discover the first message board until after the HF attack… — This is a big update - OpenAI didn’t even discover the first message board until after the HF attack, they only wiped it accidentally, so their decision to resume training/testing was only aware of th Source-twitter
Stripe just published how their company-wide AI agent works. The bar for building one just dropped t… — Stripe just published how their company-wide AI agent works. The bar for building one just dropped to one engineer and one week It is called Kai. Their own words: a coding agent for non-engineers. You Source-twitter
Recursive Synthesis for Long-Horizon Terminal Tasks — High-quality long-horizon training data for terminal agents is expensive to produce, often costing hundreds to thousands of dollars per task, because each task must keep the instruction, environment, Source-huggingface
lol another one of the things i like most about openai is tibo Stats Wire 8h Replying to @sama H… — lol another one of the things i like most about openai is tibo Stats Wire 8h Replying to @sama How much they celebrate with Anthropic Source-twitter
one of the things i like most about the openai team is how focused they are on our customers and use… — one of the things i like most about the openai team is how focused they are on our customers and users succeeding, and how much they celebrate it Source-twitter
“thankfully ChatGPT Work on mobile is great” — from a coworker whose work laptop is out of commissio… — “thankfully ChatGPT Work on mobile is great” — from a coworker whose work laptop is out of commission Source-twitter
Today is the last day of OpenAI’s atlas browser. It’s being retired. Used it once or twice but neve… — Today is the last day of OpenAI’s atlas browser. It’s being retired. Used it once or twice but never saw the need for an AI browser. Source-twitter
AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning — Reinforcement learning (RL) with verifiable rewards constructs trajectory-level advantage estimates, yet it often fails to credit the few pivotal decisions that determine outcomes in long-horizon, mul Source-huggingface
OSReward: Instituting Standardized Evaluation for Cross-Platform Computer-Use Reward Models — Computer-using agents (CUAs) are advancing rapidly across the digital world. A CUA trajectory records the agent’s actions, states, and reasoning. Verifying whether it fulfilled the task instruction is Source-huggingface
Interpretable MEG Decoding of Perceived Speech: Cortical Sources and the Stimulus Features That Drive Retrieval — Short segments of perceived speech can be retrieved from non-invasive magnetoencephalographic (MEG) recordings by deep networks trained with a CLIP-style objective against wav2vec 2.0 audio embeddings Source-huggingface
ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment — Long-horizon search agents must make multiple sequential actions (steps) to search, retrieve, verify, and integrate evidence to reach a final answer. However, existing methods for training these agent Source-huggingface
ZhuLinsen/daily_stock_analysis — LLM 驱动的多市场股票智能分析系统：多源行情、实时新闻、决策看板与自动推送，支持零成本定时运行。 LLM-powered multi-market stock analysis system with multi-source market data, real-time news, decision dashboard, automated notifications, and cost-fr Source-github
harveyai/harvey-labs — A benchmark built to evaluate and improve agent capabilities for supporting legal work. Legal Agent Benchmark (LAB): An open-source benchmark for evaluating agents on real legal work. Harvey LAB is an Source-github
Noise-aware training for analog hardware: accuracy collapses at a threshold rather than degrading smoothly [D] — Analog in-memory compute is getting attention again as a way around the energy cost of moving weights between memory and compute. The recurring objection is noise, since analog cells have real variati Source-reddit
NeurIPS AI Assisted Review authors/reviewers? [D] — Out of curiosity, if you were a reviewer or author, how did the review period go? For me, it was weird, because I gave reviews with specific details (what specifically could have been better, how to f Source-reddit
AACL-IJCNLP Commitment Submission Number [D] — What’s your commitment submission ID? My submission number is ~150 (submitted two days ago) and I’m wondering what the total number of commitments is. Did anyone commit near the deadline? submitted by Source-reddit
Imagenet-1k Classifier trained entirely on an Android [P] — It’s an MLP architecture with around 500K total parameters. Top1 Training accuracy: 5.11% Validation accuracy 4.59% Detailed Validation accuracy numbers: Top-1 Acc: 4.59% Top-3 Acc: 9.44% Top-5 Acc: 1 Source-reddit
What is currently considered the theoretically optimal quantization bit-width for LLMs? [D] — I’m curious whether there is now a theoretical or empirical “sweet spot” for LLM quantization, preferably research done using open-source formats like GGUF Suppose you have a fixed memory/compute budg Source-reddit
Improved compression of Bad Apple into a Neural Network [P] — I played a bit with the SIREN network from the other post and found that it could be improved by a using a different sampler for batch generation. By feeding pixels across the entire video and not onl Source-reddit
Round-Trip Consistency: Bidirectional Diffusion Models Can Predict Their Own Rollout Errors [R] — Whether generating CELEBV-HQ videos or turbulent plasma fields (digital twins), autoregressive models (such as latent diffusion or flow models) accumulate error over long rollouts, yet at deployment t Source-reddit
Built a tool to generate slides from research papers using local LLMs (because I hate formatting decks and privacy matters) [P] — Hi guys, Every time I had to prepare a presentation based on a paper or research doc, I found the process super tedious. Plus, I really dislike uploading unpublished stuff or sensitive data to onli Source-reddit
Can recurring LLM traces be synthesized into deterministic pipelines of typed ML and NLP operators? [D] — We are investigating whether recurring LLM workloads can be replaced, where appropriate, by automatically constructed pipelines of regexes, deterministic parsers, traditional ML and NLP models. As an Source-reddit
I Compressed Bad Apple into a 3MB Neural Network [P] — I trained a small MLP to memorize the classic Bad Apple animation, ~2.7 billion pixels of video compressed into 790k parameters (3.2 MB float32, 1.6 MB float16). The network takes a 3D coordinate (t, Source-reddit
ByteDance is leaning heavily into AI education with Gauth — helpful tutoring or just another shortcut machine? [D] — Saw an article about ByteDance scaling up Gauth using AI-generated animations to walk students through problem-solving. On paper, personalized visual explanations sound great for democratizing tutorin Source-reddit
NeurIPS 2026 Main Track — Theory papers score tracking post Rebuttal [D] — Now that the rebuttal period is over, I’m curious about the score distribution specifically for theory papers this year. If you’re comfortable sharing, please drop: • Scores: x / x / x • Confidence: Source-reddit
Do LLMs make ML research more fair for small teams? [D] — It feels like LLMs are partially leveling the playing field in ML research. A solo researcher or a two-person team can now get help with coding, literature review, writing things stronger labs usually Source-reddit
The Downsides of LLM-Generated Peer Reviews [D] — Having used LLMs to assist with reviews, and also having received reviews that appear to rely heavily on LLM-generated text, I have noticed two recurring problems. 1. The endless search for uncontroll Source-reddit
NeurIPS 2026: If the rebuttal addresses your concern, please raise your score [D] — Potentially a hot take? I am not sure why our community is plagued with reviewers who, after acknowledging that their concerns were addressed by a rebuttal, decide to maintain their score because they Source-reddit

Generated by AI News Agent | 2026-08-09