AI Reading for Tuesday February 18
Grok 3 leads in Chatbot Arena by a little bit across all categories (tied for math). - Chatbot Arena
Some coverage of the Grok 3 demo and pricing, can't use beta without an X subscription. - The Register
Wins 51% against OpenAI's top model, so you swing 1% of votes, < 100 overall, 3-4 head to head, it goes the other way, not that anyone would do that of course. But according to this benchmark it caught up to the leaders but is statistically now in the top group with OpenAI, Gemini, Claude, DeepSeek.
Maybe Sam Altman's wish that Elon Musk would compete on product was granted. Given the scrupulous honesty in all of Elon Musk's words and actions, his sense of fair play toward OpenAI, I'm sure there were no shenanigans LOL, wouldn't be crazy hard to de-anonymize using AI or a backdoor ping.
Karpathy says it's pretty pretty good - X (formerly Twitter)
I probably wouldn't want to be the industry guy or journo who says Grok 3 not all it's cracked up to be.
I think if Musk wants to really make an impact and stick it to OpenAI, has to steal a page from DeepSeek's playbook and make an excellent open source model available, let enterprises use it in Azure and AWS and GCP. The appetite to send your intellectual property to another platform is low.
I'm not sure xAI will do that immediately, hard to see how they make money and justify the billions raised if they do that. They might pitch (cough-pressure) cloud platforms for a deal similar to what OpenAI does with Azure. If customers want Grok then AWS/GCP will be very receptive to be more competitive with Azure.
Watch the video if you are into it, not very watchable to me or a lot of people, viewer count declined gradually as opposed to people virally spreading the word as it went on - X (formerly Twitter)
Free speech extremist fires engineer over a tweet - Futurism
OpenAI looks for possible governance poison pills to fend off hostile takeover from Elon Musk
Robot submarines to patrol the oceans and protect cables and infrastructure - WSJ
A $35B, 3GW AI Data Center Is Planned for South Korea - WSJ
How to learn to work with your new AI coworker - Fast Company
Google’s AI Efforts Marred by Turf Disputes - The Information
Tencent will use DeepSeek in addition to its own model - FT
10 Key Security Operations Center Challenges and How AI Helps Address Them - Hackread
How AI Is Narrowing Our View of the World.
How much expert human content and correcting AI that goes off the rails is needed to avoid an AI game of telephone and knowledge collapse in the information ecosystem?
AI-generated optical illusions can sort humans from bots - New Scientist
Ilya Sutskever's Safe Superintelligence reportedly raising $1B+ on $30B valuation - SiliconANGLE
Unapproved 'shadow AI' creates reputational risks, IP leaks, IT headaches - VentureBeat
Reddit mods struggle with AI slop, want better tools - Ars Technica
AI Companions: the future of friendship or a dangerous illusion?
Creepy men create AI girlfriends and verbally abuse them - New York Post
Sometimes talking things through with AI could help dealing with messy situations and people, but also seems likely users might learn to relate to AI companions in ways that make it harder for them to deal with real people.
Follow the latest AI headlines via SkynetAndChill.com on Bluesky