Skip to content

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

Daily almanacNº 1

Monday

November 9

New moon · 0 %Leonids · Nov 17

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

Artificial intelligence

Google introduces TPU 8tx and 8i Max to broaden its inference options

Google announced TPU 8tx, a confidential-inference system built on the same silicon as TPU 8t and paired with servers using fifth-generation AMD EPYC processors. The company says it can deliver up to twice the latency improvement over 8t with a near-zero performance penalty from confidential computing. It also introduced TPU 8i Max, which combines two 8i chips and doubles compute, SRAM, and HBM bandwidth compared with 8i.

Aster editorial team · · via Google

AI-generated image

Latest

Artificial intelligence

WorldParticle uses one model to simulate very different materials

MIT researchers developed WorldParticle, a deep-learning system that represents different kinds of matter as particles and predicts how they move. The team tested it on six physical phenomena spanning sand, soft objects, and molecular protein dynamics. MIT reports that the model was more accurate than numerical baselines and, in one protein test, generated the dynamics about 100 times faster.

Aster editorial team · · via MIT News

Artificial intelligence

FLEX helps robots infer preferences beyond a literal instruction

An MIT team introduced FLEX, a framework that uses language to turn human corrections into broader preferences for a robot. The process adds a correction, generalizes it with a language model, and masks parts that should not apply in every context. In simulations involving a robotic arm and tasks with coffee, a candle, and a bowling ball, the system used GPT-4o and took about five seconds to process each interaction.

Aster editorial team · · via MIT News

Artificial intelligence

AI's power demand drives a search for used turbines

Utilities and data-center developers are seeking used or refurbished gas turbines, along with systems adapted from aircraft engines, as new equipment faces long delivery times, Reuters reported. Orders exceeding 100 GW are competing with estimated annual manufacturing capacity of 60 to 70 GW. In the United States, the average wait for a new turbine has reached six years, according to industry data cited by the news agency.

Aster editorial team · · via Reuters

Artificial intelligence

Companies use more AI, but spending has nearly stopped growing, Ramp says

Ramp, a business-spending platform, reports that spending on “tokens” (the unit AI is billed in) rose just 1.5% between August and September, after growing 42% in July. It picked up again in October: 8.9% from September 23 to October 23. Its lead economist attributes the slowdown to price cuts and cost controls, not weaker demand. The median company spends $12 to $15 per employee per month; the top 10% spend $600 to $1,000.

Aster editorial team · · via TechCrunch

Today's games

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

See your streaks →

Hidden constellation

Stars appear one by one. Guess the constellation in six tries or fewer.

6 triesPlay

Cosmic word

A five-letter space word. Gold tiles are in place; blue ones belong elsewhere.

6 triesPlay

Which planet am I?

Every clue makes it clearer: planets, moons and stars.

5 cluesPlay

Note of the day

Listen to a short melody and guess it note by note.

6 notesPlay
Artificial intelligence

Amazon lets you shop by texting Alexa

Amazon announced that customers can text-message Alexa’s AI agent to search for products, add them to a cart and buy them. You can also send a photo of an item or an outfit to find similar listings. Alexa will learn your preferences over time to suggest things. For now it is open only to select U.S. customers, and the company says it will reach millions in the coming weeks, ahead of holiday shopping.

Aster editorial team · · via TechCrunch

Artificial intelligence

Claude used to probe a 40-year-old cell mystery, with hits and misses

An Anthropic biology team used Claude to study vault particles, hollow barrel-shaped capsules first described in 1986 whose function is still unknown. The AI helped pick candidate proteins from about 15 billion predicted ones and process microscope images, and cut a lab step from three days to 90 minutes. It also made mistakes: an additive it recommended delayed the work two weeks, and a map it produced merged the particle with its mirror image. The project took under four months.

Aster editorial team · · via The Neuron

Artificial intelligence

Independent evaluator finds Anthropic is now improving its AI twice as fast

METR, a Berkeley nonprofit, reviewed about 11 weeks of Anthropic’s internal data and interviewed its staff. It concluded that the company is improving its models’ capabilities on well-specified tasks roughly twice as fast as a year ago. It ruled out more training compute as the cause; its best explanation is that AI is increasingly helping improve AI. Still, it says AI can’t yet train models on its own.

Aster editorial team · · via AP (via ABC News)

Studio

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

Tuner

Guitar, bass, ukulele and voice.

Metronome

30 to 300 BPM, with accents.
Artificial intelligence

Apple cuts how much its employees can spend on AI each day

Apple lowered some engineers’ daily AI budget from $300 to $200 last month, according to Bloomberg. More cuts are planned: down to $100 a day, and as low as $25 for some non-technical teams. Employees who need more can ask for an exception, and some teams will move to cheaper or on-device models.

Aster editorial team · · via The Decoder

Artificial intelligence

OpenAI launches GPT-6 Cyber, a model that has already found 81 serious flaws in Firefox

OpenAI introduced GPT-6 Cyber, a model built to find and exploit security flaws in software. For now, only approved customers in its Daybreak Red program, which covers authorized security testing, can use it through the API and Codex. In testing it found 81 previously unknown serious flaws in Firefox, and Mozilla has started fixing them. OpenAI rated it a “Critical” cybersecurity risk, the highest level on its scale.

Aster editorial team · · via The Next Web

Artificial intelligence

OpenAI launches GPT-6 Cyber, a model that found 81 critical flaws in Firefox

It’s OpenAI’s model for finding software flaws, building working exploits and suggesting patches. The company says it uncovered 81 critical vulnerabilities in Firefox, which Mozilla has fixed, and in the Linux kernel it reproduced 1,724 flaws and built exploits for 948. OpenAI rates it a “Critical” cybersecurity risk and for now offers it only to select Daybreak Red customers. The figures come from the company itself, and there is no independent testing yet.

Aster editorial team · · via The Decoder

Every story in The Weekly Report No. 3 » News archive »

AI glossary

Full glossary »
AI

Language model

An AI system trained on vast amounts of text to understand and generate language.

AI

Token

The small chunk of text (a word, part of a word or a symbol) that a language model reads and writes.

AI

Hallucination

When an AI model makes up facts, quotes or details and presents them confidently as if they were true.

AI

Open weights

An AI model whose trained parameters are published so anyone can download and run it.

AI

AI agent

An AI system that does more than answer: it plans steps and uses tools to complete a task on its own.

AI

Context window

The maximum amount of text, measured in tokens, that a language model can take into account at once.

On Aster we talk about…

Back to top ↑