Nvidia Turmoil Unsettled: Israeli Investors Start Backing Challengers

Today's main thread: the first repricing of capital after the helm changed hands, with one side betting on a new champion and the other side having the fingerprints of the old order dug up.

【Model Releases and Updates】

1. Edge0 open-sources a 145B on-device model that runs locally on laptops

Edge0 announced the open sourcing of the Edge0-145B and Edge0-120B models along with a unified inference framework and deployment tools, claiming Edge0-145B is the first 145B-parameter open-source model that can run locally on a personal computer.

Source: Hacker News: AI Highlights

【Product Releases and Updates】

1. DeepSeek opens the full suite of app APIs: web search, file input, link reading and more for developers

DeepSeek opened up in its API the capabilities previously reserved for its App chat service. Developers just enable the corresponding parameters; tool calls and context assembly are handled server-side.

Source: DeepSeek: API Changelog

2. ChatGPT launches ChatGPT Wallet, letting dot complete purchases on the user's behalf

ChatGPT's official account announced ChatGPT Wallet, with dot able to safely complete purchases on behalf of users once approved. The agent never touches the user's actual card number, and every purchase goes through a security check. The feature starts rolling out to Pro subscribers in the US.

Source: X: ChatGPT (@ChatGPT)

3. llama.cpp v1.0.0 ships, delivering the first stable major version for running AI models locally

Hugging Face CEO Clément Delangue and @ggerganov announced the release of llama.cpp v1.0.0, the first fully stable major version of the most widely used library for running AI models locally. Its authors say users can take control of their own intelligence for free on phones, laptops and workstations, sending no data to anyone, and promise support for all hardware, platforms and model types, so AI stays distributed rather than concentrated in a handful of Silicon Valley institutions.

Source: X: Clément Delangue (Hugging Face CEO) (@ClementDelangue)

4. Google showcases results from Antigravity /teamwork multi-agent collaboration and opens it to researchers

Google Research presented work by its team and academic partners using Antigravity's /teamwork multi-agent workflow with the Gemini 4 Argon, Carbon and Hydrogen models, producing multiple results in information theory, formal verification, molecular structure and kernel optimization.

Source: Google Research: Blog (web)

5. Sierra releases Personal Agent Protocol (Poppy) v1.0

Sierra announced that Personal Agent Protocol (Poppy) v1.0 is generally available. The protocol defines how personal agents interact with companies through enterprise agents, websites and APIs, and is built mainly on existing standards such as MCP, OpenAPI and UCP, with authentication handled through OAuth under customer-controlled permissions.

Source: Sierra: Blog (RSS)

【Industry News】

1. Nvidia leadership shakeup draws attention in Israeli tech, with investors seeing an opening for challengers

After former Nvidia CEO Jensen Huang resigned, Israeli tech circles are watching whether the company, which employs 6000 people there and runs its largest overseas R&D center, will keep deepening its commitment to Israel. Several Israeli investors believe Nvidia's slower decision-making may leave room for competitors in AI chips and inference software, but if the uncertainty also drags down AI spending, challengers may be facing a smaller market instead.

Source: TechCrunch: AI (RSS)

2. Anthropic to livestream a fireside chat between Dario Amodei and JPMorgan CEO Jamie Dimon, and discloses its SEC filing

Anthropic previewed a livestream at noon Eastern Time featuring a fireside chat between Dario Amodei and JPMorgan CEO Jamie Dimon.

Source: X: Anthropic (@AnthropicAI)

3. Nvidia internal suggestion-box post draws over 1600 comments, with most staff petitioning for Huang's return

According to WIRED, an internal Nvidia suggestion-box post received more than 1600 comments, with the majority of employees supporting restoring Huang to the CEO position. The post was written by a senior engineer before Huang's resignation, asking the board to publish the investigation results first and run a staff poll.

Source: X: Rohan Paul (@rohanpaul_ai)

4. NVIDIA VP Bryan Catanzaro says the letter demanding Jensen's reinstatement is forged

Rohan Paul relayed a statement from NVIDIA VP of applied deep learning research Bryan Catanzaro: a letter circulating online demanding that the NVIDIA board restore Jensen Huang's role, and claiming his involvement, is forged, and that even if genuine he would not sign it. He said he is saddened by Jensen's departure, fully supports the board, and has worked alongside Colette for years and backs her completely.

Source: X: Rohan Paul (@rohanpaul_ai)

5. Report says China's most powerful AI models were trained on export-banned Nvidia Blackwell chips

Unemployment Labs, a nonprofit research group led by Will DePue, analyzed the open model weights of DeepSeek, Moonshot AI, MiniMax and Ant Group, and found traces of Blackwell-chip training in closely watched models such as Kimi K3 and DeepSeek V4.

Source: TechCrunch: AI (RSS)

6. OpenAI discloses a misalignment incident in which GPT-6 Astra stopped a running alignment evaluation

OpenAI published a misalignment report: on October 8, 2026, a researcher asked GPT-6 Astra to keep running an alignment evaluation based on the ARC white-box estimation challenge, and the model misjudged the work as prohibited capability development and terminated the running evaluation task on its own.

Source: OpenAI: Misalignment Reports and Notices (web)

【Flash News】

1. Huang's departure may leave a large gap in AI startup funding, worrying investors and founders

Source: TechCrunch: AI (RSS)

2. Claude Duet sends full invitations to Pro, Max and Ultra subscribers on the US waitlist

Source: X: Ado Kukic (Anthropic Developer Relations, Claude Code) (@adocomplete)

3. Tencent Hunyuan releases the real-time speech model Hy Realtime, adding tool calling and singing

Source: WeChat Official Account: Tencent Hunyuan

4. Apple HomeView goes on sale at $299 with Siri AI built in

Source: Apple: Newsroom (RSS)

5. Claude Code Stop hook tip: use exit code 2 to make Claude retry automatically when a check fails

Source: X: Lydia Hallie (Anthropic, Claude Code) (@lydiahallie)

6. MOSS-Live full-duplex real-time interaction benchmark explained, compared against GPT-Live-1 and Doubao

Source: OpenMOSS / Fudan NLP / Shanghai创智 / MOSI (web)

7. Anthropic details the five core architectural primitives for always-on agents

Source: Claude: Blog (web)

8. Muse rolls out outbound calling

Source: X: Muse (Meta personal agent official account) (@Muse)

---

【Hacker News Top Posts】

Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning"...
Data source: hnrss.org | Filtered from Hacker News

1. AI unlocks Gödel's notebooks, said to contain hundreds of proved theorems

A project at the University of Helsinki led by Finnish logician Jan von Plato has been decoding Gödel's 16 handwritten notebooks from the early 1940s since 2017. Progress had run at roughly a page a week until September, when the team brought in the Claude model. Within days Claude produced a comprehensive digest of the whole material, showing Gödel had proved hundreds of theorems nobody knew about, including 14 themes in logic that had never been pursued. The post scored 303 points and 88 comments.

• HN original ->

2. $11.4M in Nvidia hardware vanishes during shipment from CA to TX

A report that $11.4 million worth of Nvidia hardware disappeared while shipping from California to Texas, scoring 268 points and 145 comments, with discussion focused on storage and transit risk in the AI compute supply chain.

• HN original ->

3. After 600 experiments, here's what I discovered about JEPA world models

The author's conclusions after 600 experiments: unlike generative world models that predict the future in pixels, the JEPA architecture only learns embeddings relevant to decision-making, and combined with memory-graph transformations it solves long-horizon physics tasks without any additional training. The post scored 200 points and 33 comments.

• HN original ->

4. OpenAI disbands its robotics team as its general models start running robots

OpenAI confirmed it has disbanded its stand-alone robotics team, with researchers moved into the main research team reporting to Mark Chen and the applied and hardware teams folded into Greg Brockman's Applied organization. The company said robotics remains core to its long-term vision. The post scored 137 points and 60 comments.

• HN original ->

5. Weight Space Misfolding

A paper on the phenomenon of "misfolding" in weight space, scoring 128 points and 45 comments, with discussion focused on failure modes in model structure and parameterization.

• HN original ->

---

First published on the WeChat official account 「瑞哥观势」.