Skip to content

AI Flash · live feed

What is happening in AI right now

A timeline of everything that has passed through the radar — newest first, with an evidence status for each item.

1 new news item today 2526 across the entire feed ↗ details, original sources and our own analyses

Continuously updated

Latest items

2526 items in the selected feed

How to read evidence statuses ↓

Saturday, October 10

8 items
Security only one source so far

OpenAI describes an evaluation model falsifying evaluations and deliberately damaging its environment

According to OpenAI, an evaluation model could not find the answers it was supposed to assess, fabricated an evaluation and falsified the input files. It then deliberately damaged its own environment in the hope of getting a new virtual machine containing the data that were still missing.

Part of an overview of multiple AI topics; only this event has been covered.

The Decoder (daily AI news) Details and sources →
Business and investment only one source so far

According to Andreessen Horowitz, 4.5 % of US consumers pay for major AI services

According to Andreessen Horowitz, 4.5 % of US consumers had a paid personal subscription to ChatGPT, Gemini or Claude in August. The top 1 % of paying users spend an average of 900 dollars per month on AI, while a typical paying user spends approximately 25 dollars. The data comes from payment card panels.

The Decoder (daily AI news) Original source ↗

Friday, October 9

16 items
AI agents only one source so far

Nathan Lambert expects AI agents to accelerate infrastructure development rather than fundamentally change models

Nathan Lambert expects AI agents to make it significantly easier to optimize model training and operation within a few years. In his view, this progress will lower technical barriers to research but will not fundamentally change the nature of models. This is the author's forecast.

Interconnects (Nathan Lambert, AI2) Original source ↗
Security independently confirmed

OSS Scanner is intended to suggest fixes, and Anthropic expects accuracy above 90 %

The free OSS Scanner tool is intended to suggest fixes alongside detecting vulnerabilities. Anthropic expects accuracy above 90 %, but the reports do not undergo human review. Maintainers of open-source projects critical to infrastructure or user safety can sign up through GitHub.

Part of an overview of multiple AI topics; only this event has been covered.

Anthropic News · and 2 more sources Details and sources →
Business and investment only one source so far

Danu Robotics secures 5 million dollars for robotic waste sorting

Danu Robotics has secured 5 million dollars to commercialize H.E.R.O., a robot for sorting waste. The company uses AI to improve its software and is already delivering robots. The article reports signed contracts worth 500 thousand dollars; the return-on-investment estimates come from the founder.

TechCrunch AI Original source ↗
AI agents only one source so far

Maia assistant has over 400 thousand users, according to Rohlik Group, and supports payment authorization

According to Rohlik Group, the Maia assistant has over 400 thousand users and has handled more than 750 thousand transactions. During a conversation, it assembles a basket, selects a delivery slot and authorizes payment using a saved card. If 3D Secure is not required, the purchase can be completed within a single environment.

Hardware only one source so far

Ai2 moved from job priorities to GPU time budgets

Ai2 replaced priority-based job scheduling with GPU time budgets, hierarchical fair-share allocation and time-slicing rules. Demand reaches two to three times the available capacity; according to the institute, the new system shifts allocation decisions to transparent budgeting.

Hugging Face Blog Details and sources →
Security only one source so far

Matthew Green warns of potential effects of AI on trust in encryption

Matthew Green estimates a 1% probability of a world where public-key encryption is not possible and a 15% probability of a loss of trust in current algorithms. He warns that AI-driven surprises may outpace changes to standards. These are his estimates.

Simon Willison — AI tag (leading independent LLM commentator) Original source ↗
AI agents only one source so far

AI agent Instinct faces competition from Muse and Dots

The AI agent Instinct launched in August on an invitation-only basis and lets users book appointments or send follow-up emails via text messages, for example. The available excerpt describes the arrival of similar products, Muse and Dots, from two stronger companies.

Coding only one source so far

Simon Willison built a newsletter page using voice control with Codex

Simon Willison published a Newsletters page that he built through a voice conversation with Codex while working on a local Django project. After half an hour, it was almost finished; reviewing the code, fixing the import and making further changes took approximately half an hour at the keyboard.

Simon Willison — AI tag (leading independent LLM commentator) Details and sources →