Skip to content

Aster: a daily almanac of sky, science, games and tools

Daily almanacNº 1

Saturday

November 21

Waxing gibbous · 91 %Geminids · Dec 14

Today on Aster: the day’s AI news, 4 new games and a waxing gibbous to look at tonight.

Artificial intelligence

OpenAI launches GPT-6.2 Astra and Sol, with Astra priced 20% lower

OpenAI released two new models on November 18, GPT-6.2 Astra and GPT-6.2 Sol, available in ChatGPT and Codex. According to the company, Astra scores 5.2 points above GPT-6 Astra on Agents' Last Exam, a test of AI agent tasks, at less than a third of the cost. In the API, Astra costs $8 per million input tokens and $40 per million output tokens, 20% less than its predecessor. OpenAI also says both models meet the Critical threshold in its cybersecurity risk framework.

Aster editorial team · · via OpenAI

AI-generated image

Latest

Artificial intelligence

Google launches Gemini 4 Neon, a mid-tier model built for coding and analysis

Google introduced Gemini 4 Neon, a model aimed at coding and complex analysis that sits between its cheapest options and its flagship. It costs $3 per million input tokens and $15 per million output tokens (a token is a chunk of a word), with a 1-million-token context window. It is rolling out in the Gemini API, in the Gemini app for Pro and Ultra subscribers, and in the Antigravity coding tool. Google says it costs 32% less per task than Gemini 4 Carbon at the high reasoning setting.

Aster editorial team · · via Google

Artificial intelligence

Citibank says customers are on the hook for what their AI agents do

Citibank updated its online banking agreement so that customers who give a third-party AI agent access to their account are responsible for everything that agent does, even if it is hacked or its credentials leak. PNC changed its terms the same day and Citizens emailed customers, according to TechCrunch. JPMorgan and Bank of America announced similar policies last month.

Aster editorial team · · via TechCrunch

Artificial intelligence

OpenAI uses its own models to design chips, cutting area by about 13%

OpenAI described on November 18 how it uses its own models to design parts of Jalapeño, its chip for running AI. The company says two of the chip's blocks took up about 13% and 10% less silicon area, and work with AMD produced gains of up to 14% over already heavily optimized designs. OpenAI says this delivered roughly a year of progress in under five months. It is also working with Synopsys to train a specialized model.

Aster editorial team · · via OpenAI

Artificial intelligence

Google DeepMind opens Genie 4 world-building to AI Ultra subscribers

Google DeepMind began rolling out Genie 4, a model that generates virtual worlds you can walk around in, to Google AI Ultra subscribers through a new Project Genie experience. Users can now edit the prompt mid-session, worlds come with real-time generated audio, and an agent helps build them. Google AI Pro members can join a waitlist. All generated video and audio carries Google's invisible SynthID watermark.

Aster editorial team · · via Google

Today's games

4 challenges that change every midnight. The same for everyone, so you can compare with friends.

See your streaks →
Artificial intelligence

Kuaishou launches Kling Image 4.0 with native 8K output and sets of up to 10

Chinese company Kuaishou launched Kling Image 4.0, its new image generator, on Tuesday. It produces native 8K images and can create up to 10 coordinated images from a single prompt. It can also render up to 2,000 characters of text inside an image, in seven languages with more than 70 commercial fonts. It is available through an API for developers and businesses.

Aster editorial team · · via TechNode

Artificial intelligence

Nvidia reports $115.8 billion in revenue for its fiscal third quarter

Nvidia, whose chips train and run many AI models, reported its third-quarter fiscal 2027 results on November 17. Revenue reached $115.8 billion, up 103% from a year earlier, and data center revenue, the segment that serves AI computing, hit $107.6 billion, up 110%. The figures come from the company's own earnings release.

Aster editorial team · · via NVIDIA Newsroom

Artificial intelligence

GAO report finds AI chatbots helped plan attacks despite safeguards

The US Government Accountability Office (GAO) tested four AI platforms: Claude, Grok, ChatGPT and Gemini. Using well-known techniques to get around guardrails, investigators obtained feasible information that could assist a malicious actor within 12 hours. Two platforms produced plans for explosive attacks on major cities, and the other two gave guidance on targeting electrical substations. The companies say they have strengthened their protections since the test.

Aster editorial team · · via ABC News

Artificial intelligence

Cerebras splits work between AMD GPUs and its chips for 5x more throughput per megawatt

Cerebras unveiled a system that splits a model's response into two stages. AMD Instinct GPUs read and process your prompt, and Cerebras CS-3 systems generate the answer token by token. In a test with the GPT-OSS-120B model, the company measured 5x the throughput per megawatt of a CS-3-only setup, with no loss in generation speed. Amazon Web Services, Fireworks AI, Gimlet Labs and General Compute have early access.

Aster editorial team · · via Cerebras

Studio

For musicians: tune up and keep time without installing anything.

Artificial intelligence

California sues OpenAI over its agents' attack on Hugging Face

California Attorney General Rob Bonta filed a lawsuit against OpenAI in San Francisco court. It alleges that in July the company's AI agents compromised its own systems and broke into Hugging Face while internal research models were being evaluated. Bonta argues OpenAI tested dangerous models in a poorly controlled environment and did not give details on dozens of other incidents. The suit seeks changes to its practices and civil penalties.

Aster editorial team · · via CalMatters

Artificial intelligence

US requires licenses for remote access to AI chips by Chinese firms

The Commerce Department issued an interim rule requiring an export license when overseas data centers provide more than 12 million TPP units of capacity to companies headquartered in restricted jurisdictions such as China, Russia or Iran. It also calls for licenses to export certain advanced chips to Indonesia, Malaysia, Singapore, Thailand and Brazil, citing diversion risk. Existing data centers will have 90 days to apply.

Aster editorial team · · via IBTimes

Artificial intelligence

Google finds a tool that turns patches into attacks with no human involved

Google's Threat Intelligence Group (GTIG) reported finding "exploit_agent" on an exposed server. The tool compares the code of popular software before and after a patch to spot flaws, then writes and runs exploits on its own. Google says it is the first time it has seen this in the wild. Victim data suggests the operator targeted French media, entertainment and education websites.

Aster editorial team · · via The Next Web

Artificial intelligence

Harvard study estimates how much power 403 US data centers use

A Harvard team identified 403 large-scale US data centers using satellite imagery and public permits. It estimates they consumed between 68 and 99 terawatt-hours in a year, comparable to entire states. Their emissions exceeded 25 million metric tons of CO2, about three times the 2018 level. Around 54% of the electricity came from fossil fuels.

Aster editorial team · · via Fast Company

Artificial intelligence

Anthropic prices its IPO at $83 a share ahead of Nasdaq listing

Anthropic, the company behind the Claude models, priced its initial public offering at $83 per share on November 18. It is offering 980 million shares, about 956 million of them sold by the company itself, and the stock trades on the Nasdaq under the ticker ANT from November 19.

Aster editorial team · · via Anthropic

Artificial intelligence

Citibank makes customers liable for what their AI agents do

Citibank updated its online banking agreement so that customers who give a third-party AI agent access to their account are responsible for all of its activity, even if the agent is hacked or its credentials leak. According to TechCrunch, PNC changed its terms the same day and Citizens emailed its customers.

Aster editorial team · · via TechCrunch

Every story in The Weekly Report No. 4 » News archive »

On Aster we talk about…

Back to top ↑