All AI news

AITibor Blaho says models this capable still make stupid mistakes, even in auto mode, though he still uses Opus and Fable together with Sol and Astra. The quoted post reports that Claude Fable 5.1 max accidentally deleted seven repositories and a full day of unpushed work while in auto mode.

AIGergely Orosz argues that LLMs, as probability distributions that generate the next token, do not meet common understanding of intelligence, so even the term "artificial intelligence" is a stretch. He says hallucination is a feature of this design rather than a bug, and criticizes renaming LLMs as "superintelligence." Simon Willison's reply calls the "Super Intelligence" label stupid.

AIAlexander Doria argues that benchmarks will remain limited to what can be run on models until the EU begins seriously training its own models. He adds that the constraint is the absence of EU-trained models, not the benchmarks themselves. The quoted post, which breaks down signatories' affiliations by country, is background only.

AIAether AI's CRIS-0, its first causal robotic intelligence system, operates in a real home and models how actions change the physical world. Per the post, its causal world model predicts how conditions could change under different robot actions, while a causal agent keeps task context and selects capabilities at each stage. A unified tool interface connects navigation, learned action models, rule-based functions, and result checks.

AIMeliá Hotels International, a hotel chain with more than 380 hotels across four continents, is cited in an AWS post as having taken a step the post describes as the approach needed. The source gives no further details on the specific technology, implementation, or results.

AIA company used AWS to migrate its entire COBOL-based central reservation system to microservices. The migration handles 50 million daily availability requests, up from 26 million, with faster response times.

AIAn AWS customer reports that new features now ship in one month instead of four, with 60% compute cost savings worth seven figures. The post also cites a 75% faster time to market, near 99.99% availability, and a four-year project finished in two years.

AIAether AI, founded by UCSD assistant professor Biwei Huang, has released official demos of its CRIS-0 causal intelligence system for robots. In tests, the robot recovered from external disturbances in 9 of 10 random trials, typically within about 2 seconds, and stopped within 0.2 seconds when a human hand entered the workspace during a microwave-door task.

AILenovo's TianxiCode, paired with DeepSeek-v4.1-Flash, ranked first on the SWE-bench-Live Lite leaderboard with a 71% issue resolution rate and passed official Verified review. The framework combines multi-hop retrieval, autonomous planning with multi-turn tool calling, and test-driven self-correction, and will be applied to Lenovo AI hardware products.

AIAnthropic's Claude Science mapped the entire sky in ultraviolet light for the first time. AI agents downloaded data from multiple space missions, calibrated it, merged it, and used inpainting to fill gaps left by NASA's GALEX mission. In tests, predictions averaged about a ten percent deviation from actual measurements.

AIOpenAI says an internal investigation found Jasmine Wang, Tomek Korbak and Mikita Balesni committed a significant breach of trust by violating policies on handling sensitive information. The company denies the dismissals were about the researchers speaking out on AI safety, responding to an open letter in which the group said it was fired for raising safety concerns. OpenAI says the investigation found breaches beyond those in the letter but has not provided details.

AIGoogle says gemini-3.7-flash is deprecated and replaced by gemini-3.8-flash, and gemini-3.5-flash is deprecated and replaced by gemini-3.6-flash, with requests to the old strings automatically routed to the new ones. Developers should update their model strings, with gemini-3.8-flash recommended for the 3.5 Flash replacement as well. Google also says the deep-research-pro-preview-12-2025 agent will shut down on October 23, 2026, and that developers should migrate to deep-research-preview-04-2026 or deep-research-max-preview-04-2026.

AIAnthropic released Claude Haiku 5.5, its cheapest and fastest small model, which it says costs around 75% less to run than Claude Haiku 4.5. The author argues Anthropic is the only frontier model builder, though the piece also covers Nous Research's $90 million Series B and OpenAI's revenue discrepancy.

AISophos is using OpenAI's Daybreak to cut cyber-threat investigation time by 96% and automate 52% of its MDR cases. The source says the approach preserves human oversight.

AIThe article argues that AI refusal, the main safety mechanism in modern models, is unreliable and hard to draw lines for. It cites jailbreaks, classifier stacks, and studies showing refusal skewed toward repressive governments. It warns that governments and companies could use refusal to censor speech, and that refusal behavior remains poorly understood.

AISophos says agents built through OpenAI Daybreak, combined with its cybersecurity expertise, cut average response time from 38 minutes to 89 seconds for cases handled by those agents. The company says the agents help its MDR team investigate threats faster and protect customers at scale while keeping human judgment central.

AIHugging Face's Julien Chaumond quips "niche of 1 lol" in response to a post describing a hard-to-hire role. The quoted post lists the traits sought: chronically online, taste and creative ability, AI and technical understanding, and strong execution.

AIBen Tossell asks whether people genuinely believe code maintenance cannot be solved with an agent or service. The post is a short question with no further details, figures, or products named.

AIOpenBMB reports that its MiniCPM5-2B model runs at 37 tokens per second on an iPhone Air. The post presents this as evidence that small open multimodal models can run on mobile devices without a cloud GPU, with NobodyWho noting the model is available in its Chat app.

AIGoogle Cloud has launched the Gemini agent, a cloud-hosted agent that answers questions, does knowledge work, creates media, and writes and runs code from one prompt box and one API. It orchestrates across Gemini and Anthropic Claude models today, and its project-level spend caps pause an agent when a limit is reached. The article reports no benchmarks, pricing, or general availability date yet.

AIUnderdog has released Saluki 27B under Apache 2.0, a 2-bit GGUF of Qwen3.8-27B that fits in 7.89 GB, versus 54 GB for the full BF16 model. On Underdog Bench, Saluki scores 88 against 84 for the full model, and it raises parallel tool-call accuracy to 42 from 35. It runs on stock llama.cpp, but math and reasoning drop sharply, with AIME 2025 at 79.2 versus 96.7.

AIOpenAI told investors it expects $50bn in revenue this year, about $20bn less than the $70bn it had signalled last month. The gap stems partly from comparing with Anthropic, which counts revenue sold through cloud partners such as AWS and Google Cloud, while OpenAI does not. The news weighed on US tech stocks, and OpenAI is in early talks to raise $30bn at a valuation of about $1.4tn.

AIOpenAI exposed a Russian and an Iranian influence operation and banned the ChatGPT accounts involved, both of which planted content in legitimate media using fake identities. The Iranian operation, "Bogus Bylines," used seven fake journalists to place nearly 100 articles about the US-Iran conflict, while the Russian "Dark Clark" operation triggered fact-checks and official denials in Ecuador and Peru. Both operations used AI mainly for internal reporting and adapting propaganda to different languages.

AITencent has open-sourced Youtu-Parsing-Omni, a 5B-parameter omni-modal model that outputs a single structured JSON covering layout, text, tables, formulas, ASR, OCR, and video segments. It scores 96.96 Overall on OmniDocBench, the highest among the compared models, and ships with weights on Hugging Face, a vLLM plugin, and inference examples.

AINVIDIA published the Agile One S SSD Pick deployment model, GR00T N1.7 checkpoint 40000, on Hugging Face for SSD pickup tasks. The repository includes five ONNX graphs with external tensor files and two existing TensorRT BF16 engines, with original configurations and build metadata, but no retraining or re-export was performed. The files are not a robot deployment or safety qualification, and engine compatibility depends on the target GPU and TensorRT environment.

AINVIDIA has released a deployment model for Agile One S SSD pickup, based on GR00T N1.7 checkpoint 58000 and using three cameras: ego, left wrist, and right wrist. The repository republishes ONNX graphs, external tensor files, and two existing TensorRT BF16 engines without retraining or re-export, and the original export reported a numerical warning that full FP32, node, and BF16 parity did not pass all tolerances. The files are not a certified robot deployment or safety qualification.

AINVIDIA published the Agile One S Walk GR00T N2 checkpoint 1680, a walking deployment model with four cameras, on Hugging Face. The repository includes ONNX graphs, TensorRT BF16 plans/engines, and the original checkpoint files, republished without retraining or re-export. The shared Cosmos-Reason1-7B dependency and the Isaac/GR00T runtime must be set up separately, and the files are not a robot safety qualification.

AINVIDIA published the nvidia/agile_one_s_place_ssd_n17_24050 repository on Hugging Face, containing a GR00T N1.7 checkpoint 24050 model for placing an SSD with three cameras. The repository includes five ONNX graphs with external tensor files and two existing TensorRT BF16 engines, republished without retraining, re-export, or engine rebuild. Engine compatibility depends on the target GPU and TensorRT environment, and the files are not a robot deployment or safety qualification.

AIHarrison Chase's post is titled "eval driven development," presenting evals as a development approach. The main post gives no further detail beyond the title. The quoted context from Jerry Liu argues that most tasks can be solved by defining an eval and hillclimbing over it rather than hand-building a deterministic or agentic workflow.

AIHarrison Chase argues that eval-driven development works for narrowly scoped tasks but breaks down for more autonomous agents, invoking Goodhart's Law that a measure ceases to be useful once it becomes a target. He asks how such agents can be hill-climbed, and the post does not provide an answer.

AIOpenAI's GPT-6.1 Sol Ultrafast is rolling out in the API, Codex, and ChatGPT Work, running up to 8x faster than Sol Standard at $12/$60 per million tokens. StepFun's Step 5 Preview, a 600B-parameter MoE model with 27B active parameters, is now on OpenRouter, and JetBrains released the open 12B MoE coding model Mellum2.1 under Apache 2.0.

AIOpenAI has made GPT-6 available to all ChatGPT free and paid users worldwide, replacing GPT-5.6 SOL and GPT-5.6 LUNA. The model integrates Astra safety improvements that strengthen safeguards against misuse in cyber, biological, and violent domains.

AIMagnific asks X users to drop their portraits in replies to join its Magnific One breakfast club. The background post promotes Magnific One as a way to bring art direction into image generation, with unlimited use in Magnific Desktop until October 15.