AI Reading for Sunday January 26
How DeepSeek overcame US sanctions - MIT Technology Review
Huggingface published a DeepSeek-R1 how-to on reproducing their training pipeline. - GitHub
Just don't ask DeepSeek about Tienanmen - Reddit
Or Taiwan. - Reddit
o1-preview released in September, so in this instance the gap between the state of the art proprietary model and open source shrank to about 4 months, and they don’t have the successor shipped so there is a competitor that is 1/10 the price.
OpenAI had a year or more as the only game in town with GPT-3.5 and GPT-4, then they were 6 months ahead for a while, based on DeepSeek and Google's Flash reasoning, down to 3-4 months and now they are one of several at the top.
Maybe OpenAI really is the new Netscape, Google is the new Microsoft, Nvidia is the new Cisco.
Musk-Altman feud reaches White House. - WSJ
OpenAI adds o1 coding, HTML rendering to Canvas - X (formerly Twitter)
Willison says it has potential but doesn’t work great yet. - Simon Willison
Zuck says he's got a big one, 2GW data center half the size of Manhattan - The Verge
On the one hand you have the case of the missing mega models, Opus, GPT-5, relatively modest improvement in Gemini 2.0, and the rise of smaller and open source models and edge models, suggesting maybe you don’t need as much compute in the future.
On the other hand, there’s the ‘$500b’ announcement, $300b AI capex at hyperscalers in 2025, Amodei and Altman saying they will have something AGI-like soon, give us hundreds of billions or trillions and we’ll compress 100 years of human scientific progress into 10 years.
What different people said about AI at Davos - WEF
Looking for a new AI paradigm, maybe one that can learn on the fly and have a proper world model. - TechCrunch
Anthropic's Citations is like integrated RAG, improves accuracy and interpretability - Inc.
Anthropic building doc upload w/citations i.e. RAG directly into Claude now. - Ars Technica
The Apple exec tasked with whipping Siri into shape with AI - Bloomberg
An AI 'exocortex' for scientists? - TechRadar
Is AI Creating a Generation of Illiterate Programmers? - N’s Blog
Can you imagine a time when we didn't even have the Web, had to read books like K&R and the Camel Book and know them like the back of our hand?
Should an AI assistant be able to prescribe drugs? - Futurism
Banks fire up coding assistants as AI costs plummet - CIO Dive
Someone made a fake AI expert to scam journalists into using it as a source. - Fast Company
Cops Are Still Bypassing Facial Recognition Controls To Build Cases Based On Bad Matches. - Techdirt
For high-stakes stuff, you really have to use AI as part of a structured process with guardrails. AI gives an answer, people will rely on it.
Uber CEO is asked what will happen to Uber drivers when robotaxis arrive, says will take 10 years, will have both in tandem for a long time - Business Insider
The Brutalist's AI controversy - Vanity Fair
All fake, including the behind-the-scenes footage - YouTube
Follow the latest AI headlines via SkynetAndChill.com on Bluesky