Google's AI Overview stumble Google published a blog post giving the obvious explanation for those viral weird answers to AI questions last week. As I wrote, it's not clear to me how much we'll just get used to this ('LLMs are only good for some use cases' and we learn that), how much it invalidates the concept of using LLMs for general search, and how much this was a Google-specific execution failure. LINK Elon enters the model wars Elon Musk's xAI raised $6bn at a $24bn valuation to build Yet Another LLM. There are so many $25bn LLM startups now that we're starting to get to real money. Apparently this will be the one that pursues 'truth', prevents AGI from happening, builds AGI, and probably a few other things. We will see - Musk, to put it as neutrally as possible, does not have a good track record in predictions about AI. Meanwhile the company is apparently in talks with Oracle for a $10bn AI training cloud (full of Nvidia chips, of course - maybe it would be quicker just to wire Jenson the money?). NEWS, ANNOUNCEMENT, ORACLE Everything is hacked Both Ticketmaster and Santander, the European retail bank, were hacked in the last week, perhaps by the same actor, while Hudson Rock claims that this is linked to a breach at Snowflake (cloud storage). SANTANDER, TICKETMASTER The week in AI OpenAI said that it kicked some foreign government misinformation campaigns off its systems. This is also an implicit dig at open source. LINK Vox and The Atlantic signed AI training data deals with OpenAI. I'm curious about this one - there's not much real-time information (especially for The Atlantic) and yet is the archive big enough to move the needle? LINK Nvidia has a huge moat, and so, very predictable, there's now a group to try to create an open industry standard for AI datacenter chips in competition. LINK OpenAI signed PwC as a go-to-market partner. There's a tension in here somewhere - if ChatGPT is so powerful and its utility so obvious, why would you need a systems integrator to turn it into products for actual real enterprises? LINK Sony Pictures said it out loud: Hollywood studios will use generative AI to cut production costs (even though Hollywood just went through a strike that was partly about this). LINK Crypto scams Coinbase, Meta and a bunch of tech companies announced a group to try to do something about the wave of scammers ('pig butchering' etc) that are using social media to find targets and crypto for easy payment to boiler-rooms in Myanmar or west Africa. LINK Google search leaks Someone leaked a couple of thousand internal Google Search technical documents. Very exciting for SEO geeks - not yet clear if there's anything earth-shaking for the rest of us? LINK Google buys Magic Leap? Close to a decade ago, Magic Leap got a lot of attention (and a small investment from my then-employer, Andreessen Horowitz) for a great AR proof-of-concept that worked in the lab, but it was never able to scale that into a practical, wearable device. Now Google has done a 'strategic deal' (that sounds like a quasi-acquisition). Google seems to have pulled back from VR and AR in the last few years, but this signals continuing intent. Meanwhile, Apple's decision to launch the Vision Pro as a $3500 developer kit (and not to wait for AR optics) shows just how far this space still is from mass-market products with real traction. Also - a year after the iPhone or iPad launched, everyone wanted to know what Apple would announce for them at WWDC. This year, WWDC (next Monday) is clearly all about generative AI, and no-one is talking about what might come to the Vision Pro. LINK | 
No comments:
Post a Comment