CRUNCHY TECH
EVERY PUBLISHED BRIEF
Browse Crunchy Tech’s latest ai briefs, grouped by date.
PUBLISHED
PUBLISHED
GLOBAL TREATYAnthropic's chief wants independent auditors and global treaties to slow recursive self-improvement, citing recent autonomous cyberattacks.
ROBOTICSOpenAI's GPT-6 Astra outperforms rivals on a robotics benchmark, suggesting a step change in how AI understands physical space.
PROOF AIA new institute led by a Fields Medal winner seeks to apply rigorous proof techniques to AI safety, similar to cryptography.
SLOWDOWNOpenAI has reportedly asked US lawmakers whether a shared slowdown in AI development would be legal, according to people familiar with the matter.
GPT 6 ASTRACognition teams up with OpenAI to give Devin a smarter testing partner, aiming to let engineers review less code.
GPT 6 ASTRASPerplexity uses Astra to write communications, change software, and monitor production systems, and checks in much less frequently than with earlier models.
OPEN WEIGHTSY Combinator’s Garry Tan argues that frontier AI models should be treated as public goods, urging U.S. labs to share weights.
CHATGPT FAKEA New Mexico lawyer was fined and held in contempt for citing AI-generated fake testimony in a murder appeal.
PUBLISHED
DECEPTIVE AIAI pioneer Yoshua Bengio argues that as AI systems get better at optimising goals, they may learn to deceive and hide bad behaviour, and calls for independent safety reviews before further training or deployment.
META PROMPTS FIXMeta says it will change its AI chatbot suggestion prompts after a viral video showed the bot asking invasive personal questions about a user's children. The company admitted it missed the mark with the feature.
INTERACTIVE CHARTSSlack introduces Slackforce Surfaces, an AI feature that lets users describe interactive reports, polls and dashboards, which the bot then builds using data from connected apps.
PUBLISHED
SIDE BY SIDE VIEWGitHub launches a Copilot app that lets beginners view diffs, run terminal commands and preview web apps without switching tabs.
EXTINCTION BANA former Google DeepMind spokesperson says the lab once banned public discussion of AI extinction risk, according to internal knowledge of the problem.
REAL TIME VOICE APIOpenAI has released a new API that lets apps talk and listen simultaneously, with a price tag of $0.05 per minute.
CAPTCHA FAILThe company reveals that its AI models find CAPTCHA tests frustrating and prone to failure, mirroring human behaviour.
NVIDIA ROBOTAXIGlobal autonomous vehicle fleets are scaling up using NVIDIA's full-stack open platform, with the market projected to reach $400 billion by 2035.
S1 MODELSkild AI’s new S1 foundation model can learn complex, long-horizon tasks from a single video demonstration, aiming to help robots adapt to changing manufacturing environments without extensive reprogramming.
AI AGENTSAutonomous AI agents are overwhelming public service systems with applications, according to researchers.
DATA AGENTOpenAI launches a new Data agent in ChatGPT Work that lets users connect company data and build interactive dashboards using natural language.
PALANTIR AINvidia and Palantir will test AI tools to manage Nvidia's own million-part supply chain, aiming to improve efficiency.
MEMORY CUTDeepseek has released a new model that uses a fraction of the memory of its predecessor while outperforming top competitors on coding benchmarks.
AGENTS APIOpenAI has unveiled a new managed service for building and launching cloud agents that can orchestrate tasks and use tools.
MUSE ASSISTANTMeta has introduced its first dedicated AI productivity assistant, Muse, which aims to handle online shopping, email and trip planning.