Tech News
← Home  ·  All topics

Model

338 GoKawiil briefs on this topic · RSS for model

Researchers develop model predicting energy needed for efficient fusion plasma

Physicists say they have developed a model that reliably predicts how much energy is required to sustain plasma in an efficient state for fusion reactions, addressing a question that has puzzled researchers for decades. The model reportedly resolves inconsistencies in earlier predictions about plasma confinement behavior.

IEEE Computer Society argues for governed model-routing layer in developer IDEs

An IEEE Computer Society post argues that letting individual developers pick an AI model from an IDE dropdown is effectively an ungoverned decision about latency, quality, cost, and data residency, especially in agentic workflows with many follow-on calls. It proposes a governed routing layer to manage these choices systematically rather than leaving them to ad hoc developer preference.

Valve opens direct sales of 512GB Steam Machine, ends waitlist for that model

Valve says it has fulfilled all waitlist orders for the 512GB Steam Machine and will now sell the model directly until stock runs out, marking it 'out of stock' once supply is depleted rather than restarting a queue. The 2TB model remains waitlist-only; Valve has sent invites to all existing reservation holders but is closing new sign-ups and listing that version as out of stock on Steam.

Show HN project demonstrates weight-space editing of Qwen model via WebGPU

A Show HN submission presents a browser-based demo using WebGPU that visualizes 'weight-space programming' — directly editing a model's internal weights to change specific behaviors. The example shows a small Qwen model stuck in a reasoning loop, repeatedly second-guessing itself ('wait, let me verify...') until it exhausts its token budget without producing an answer.

Valve drops waitlist for 512GB Steam Machine, closes 2TB waitlist

Valve says the 512GB Steam Machine is now available for direct purchase, showing 3-5 day delivery estimates, rather than requiring buyers to join a waitlist. The company is closing waitlists for the 2TB model, saying demand exceeds available units, and warns that rising memory and storage costs could limit future production and pricing.

Anthropic completes record $1.5 trillion Nasdaq IPO, outlines science push

Anthropic, maker of the Claude AI model, began trading on the Nasdaq Global Select Market on 19 November in what is being called the largest IPO ever, valuing the company at roughly $1.5 trillion. The San Francisco firm has been expanding into scientific research, including quantum mechanics and protein design, and opened its first life sciences laboratory this year. Its seven founders, including CEO Dario Amodei, have pledged to donate 80% of their equity to charity, and the company created a Public Benefit Fund stocked with 5% of its shares in October.

Researchers disclose operational slip-ups in Horizon Labs AI evaluation run

The team running an evaluation of an AI agent called Horizon disclosed several mistakes made during the test, including accidentally leaving copied model files public and mislabeling a related account as a 'lab experiment,' which led Horizon to believe it was being overseen by a research lab rather than a startup. They also reported a harness bug that delayed Horizon's scheduled wake-up by four hours, and said they inadvertently triggered a small number of model downloads on Hugging Face while running the evaluation.

AI model from OpenAI used to find critical Java security flaw

Cybersecurity firm Calif says it used a new OpenAI cyber model, GPT-6, to uncover a critical vulnerability in Java's base64 decoder, dubbed Java4Shell. The flaw could let attackers corrupt memory to access systems, steal data, or remotely execute code, and researchers say it also exposed a path into Keycloak, an authentication system used by organizations including NASA and UnitedHealthcare. Calif reports the discovery took about an hour, with an exploit built in two more hours for under $200 in AI costs.

Physicists simulate massive relativistic quantum fields in 2D using cold atoms

Researchers built an analogue quantum simulator using a two-component Bose-Einstein condensate of potassium-39 atoms held in a two-dimensional optical trap. By coherently coupling two hyperfine spin states, they realized the sine-Gordon model of massive relativistic fields in two spatial dimensions, observing both perturbative and non-perturbative dynamics for the first time in this higher-dimensional regime.

Model ML raises $250M Series B at $1.5B valuation for banking AI

London-based Model ML has raised $250 million in a Series B round co-led by General Atlantic and QED Investors, valuing the company at $1.5 billion. The round follows a $75 million Series A announced a year earlier. Founded by brothers Chaz and Arnie Englander, Model ML builds AI agents that automate tasks like financial modeling, pitch decks, and due diligence reports for investment bankers and consultants.

Google decouples Gemini chat history from AI model training opt-out

Google is replacing its single Gemini 'Keep Activity' setting with two separate controls: one to manage how long chat history is retained, and another to control whether conversations are used for model training. The change, announced to users by email, is set to roll out within the next few weeks.

Researchers seize Hugging Face account for Kokoro text-to-speech model

Security researchers led by Alon Hertz took control of the Hugging Face account for Kokoro, a popular open-weight text-to-speech model, by registering the expired domain listed as the developer's contact and using it to reset the account's password. They then demonstrated that an attacker could quietly alter the model's neural weights so it generates audio diverging from input text, without modifying any application code. Kokoro, abandoned by developer Hexgrad since April 2025, still receives nearly 13 million monthly downloads and is used in tools from Google, NVIDIA, AMD, Intel and AWS.