AI Research: recent coverage

Papers, breakthroughs, and techniques from the AI research frontier.

• Fable 5 sits at the top of KernelBench. Jack Clark calls it “the start of a RSI loop”reddit · Jul 6, 2026
• Import AI 464: Fables writes GPU kernels; AI automation; and analog computationrss · Jul 6, 2026
• Kyutai's Pocket TTS clones a voice from 5 seconds of audio, on CPU, under MIT. Benchmarked against Kokoro, Supertonic, and Inflect-Nano for Eng. TTSreddit · Jul 6, 2026
• Does code cleanliness affect coding agents? A controlled minimal-pair studyhackernews · Jul 5, 2026
• Qwen & Gemma on deadlock situation (For Benchmarks Numbers)?reddit · Jul 6, 2026
• One day AI will be aligned enough to handle blind spots like this smoothly, and this guy is helping push us in that direction.reddit · Jul 3, 2026
• A sociotechnical threat model for AI-driven smart home deviceshackernews · Jul 5, 2026
• "An entirely too simplistic notion of intelligence has dominated both popular and scientific thought, and that this notion is, in part, responsible fobluesky · Jul 5, 2026
• Deepseek drops another HUGE breakthrough - DSpark. Waaay faster than MTP [Video explaining it]reddit · Jul 3, 2026
• Both halves of the problem are basically intractable. I’m confident that, short of specific cryptographic watermarks in the generated text, there’s nobluesky · Jul 5, 2026
• Before you blame the model, look at the conversation.reddit · Jul 6, 2026
• google/tabfm-1.0.0reddit · Jul 4, 2026
• Shin-Sekai: The Japanese Macaque Shogunatereddit · Jul 5, 2026
• Another #ContraAI🤖 book I am reading not because I want to, but because I have to: we have an obligation - as anthropologists, educators & parents - bluesky · Jul 5, 2026
• <15B local models to test on medical benchmark?reddit · Jul 6, 2026
• possible evidence of literal prompt injection by anthropicreddit · Jul 4, 2026
• People are able to detect AI generated text 75% of timereddit · Jul 5, 2026
• Follow-up: DeepSeek V4 Flash on 2x RTX PRO 6000 finishes real coding tasks faster than Sonnet and Opus, at about Sonnet qualityreddit · Jul 3, 2026
• I developed a 270 million parameter language model entirely from scratch as an independent research projectreddit · Jul 5, 2026
• What tells does AI have in non-English languages?reddit · Jul 6, 2026
• Whitepaper: The judgment problem in AI-generated API test suiteshackernews · Jul 6, 2026
• In 2022, experts predicted AI will be able to write publishable math theorems by around 2050 and win the Putnam exam around 2033. LLMs did both THIS YEAR!!!reddit · Jul 3, 2026
• Portugal just released their own LLM Amalia (9B)!reddit · Jul 3, 2026
• At work we are trying to rough out a process for how to verify and validate LLM alignment results for eventual use in analysis, and one of my co-workebluesky · Jul 4, 2026
• Emily Bender Sets the Record Straight on "Stochastic Parrots"hackernews · Jul 6, 2026
• TRACE: open-source hierarchical memory for LLM agents, 82.5% on MemoryAgentBench’s EventQA using gpt-oss-20B [P]reddit · Jul 6, 2026
• Does Speaking to Agents Like Cavemen Save 65% of Tokens? We Testhackernews · Jul 6, 2026
• The circuit that lets your brain think and seehackernews · Jul 3, 2026
• Should AI be able to prove what it knew at the time?reddit · Jul 6, 2026
• big research orgs (OpenAI, Meta's thing before LeCunn left, Ant) from anticipating anything beyond scaling transformers at any costs, which is a big bbluesky · Jul 5, 2026
• Arrived in Genève for the Global Dialogue on AI Governance, where the report of the Independent International Scientific Panel on Artificial Intelligebluesky · Jul 5, 2026
• Show HN: Open Science, open-source alternative to Claude Sciencehackernews · Jul 6, 2026
• Llama-Server is Throwing Away Your Perfectly Good KV Caches, and How to Fix Itreddit · Jul 6, 2026
• FlowerBench: Benchmarking AI Agents on Real Enterprise Workhackernews · Jul 6, 2026
• Ask HN: Are there good security benchmarks for LLMs?hackernews · Jul 6, 2026
• Alphazero go bots are superhuman, but the adversarial attacks reveal that their “understanding“ of the game is alien and imperfect. Does the same thinbluesky · Jul 4, 2026
• Context graphs: how AI agents can store and use past decisionshackernews · Jul 5, 2026
• Is AI ruining our skills? Early results are in—and they’re not good Reliance on artificial-intelligence tools degrades the abilities of physicians andbluesky · Jul 5, 2026
• When an artificial intelligence memorizes your phone number, your address, or your medical history during training, deleting that information later tubluesky · Jul 4, 2026
• Is Intrinsic Motivation a Viable PhD Topic in 2026? [D]reddit · Jul 5, 2026