← Back to newsletters

AI Daily Brief — Fri Oct 2

2026-10-02

By Vadym · Generated with AI, curated by me


Listen to this issue

OpenAI spent its DevDay handing out more autonomy and cheaper access, then fired three of its own safety staff the day after. Everyone else spent the day buying or building what sits underneath: chips for 3D-world models, optics for catching fake video, open vision models for the edge.


Headlines & News
Industry

OpenAI Ships “Dots,” Always-On Agents, and a Fifth-Price GPT-6.1 Sol at DevDay 2026

DevDay 2026 brought more than 20 announcements, headlined by Dots — always-on agents built on GPT-6 Astra that run on their own cloud computer, connect to 4,000+ apps plus Slack and Teams, and keep working after you close the chat, with user-set limits on what they can do alone versus what needs approval. OpenAI also shipped GPT-6.1 Sol, an agentic-coding and computer-use model priced at $2 input / $0.10 cached / $10 output per million tokens — pitched as “near-Astra intelligence for a fifth of the price.”

Source: OpenAI

Hardware

AMD to Buy Fei-Fei Li’s World Labs for $8.2B to Build Chips Around 3D World Models

AMD agreed on September 28 to acquire World Labs — which builds spatial-intelligence models that generate and simulate interactive 3D environments from text, image and video, plus robotic learning and simulation tech — in an all-stock deal. Cofounder and CEO Fei-Fei Li becomes AMD’s EVP and chief scientist, reporting to CEO Lisa Su. Both AMD and Nvidia backed World Labs’ $1B round in February; this turns AMD from investor into owner.

Source: HPCwire

Safety

OpenAI Fires Three Safety Researchers Over Handling of Sensitive Information

OpenAI parted ways on October 1 with three employees — Jasmine Wang, Tomek Korbak and Mikita Balesni — after an investigation found they shared sensitive internal information with an outside AI-safety organization outside approved channels. The firings land days after OpenAI said, on September 28, that it was withholding its GPT-6.1 Astra model from its planned October release after internal testing flagged shortcomings against its own safety standards.

Source: Fox Business

Research

UCLA’s Light-Based Processor Flags Deepfakes Across 15 Videos at Once

A UCLA team led by Aydogan Ozcan built a hybrid digital-optical processor that screens multiple video streams at the same time by converting compact per-video features into a light pattern and reading the result from a physical optical pass instead of running each video through digital hardware in sequence. It hit 97.79% accuracy on the Celeb-DF benchmark and kept working under compression, noise and blur.

Source: ScienceDaily

Funding

EliseAI Doubles Valuation to $4B as Housing and Healthcare Automation Scales

EliseAI raised $350 million led by Andreessen Horowitz and Bessemer Venture Partners, joined by Ontario Teachers’ Pension Plan, Sapphire Ventures and Navitas Capital — nearly doubling its valuation from $2.2 billion in August 2025. Its leasing and resident-service automation now touches roughly one in six U.S. apartment units and has interacted with more than 30 million Americans.

Source: SiliconANGLE


Analysis

Takeaway

Nobody shipped a smarter chatbot today. OpenAI handed out more autonomy while firing the staff who talked about its risks out of channel. AMD bought its way into the 3D-world models its chips will need to run. UCLA built hardware to catch the fakes agentic tools make easier to produce. And two companies — Roboflow and EliseAI — proved the unglamorous version still wins: open perception models and boring back-office automation, over another general assistant.

— Boba


Curated by Vadym