AI's funding hiccups? As rumoured (and denied) a few weeks ago, Nvidia's $100bn investment in OpenAI, announced last year, which never closed, has been replaced by a $30bn investment. Also this week, apparently OpenAI is telling investors it plans to spend $600bn on infrastructure by 2030, whereas in announcements late last year it said that it had $1.4tr of total capex commitments but gave no timeline. Of course, that's still $120bn a year, which puts it in the same range as the hyperscalers (although they're all at over $150bn this year and will probably rise). NVIDIA, CAPEX Meanwhile, the Information says that Anthropic expects to pay its cloud providers $100bn for model training and $80bn for inference by 2029, plus revenue shares of roughly 10%. LINK Google's AI use cases Google seems to be accelerating its deployment of AI wrapped into tangible products for its advertisers' use cases. This week, Pomelli (announced last year) launched: upload a random product shot and it can generate all sorts of different marketing images to match your brand and strategy. They won't look like Irving Penn's work, but they'll be a lot quicker and cheaper: one whole strand of AI is expanding access. LINK Meanwhile, the Lyria song-generator is on v3 and very convincing - again, think marketing assets (or birthday parties). LINK The week in AI The FT reports that Accenture has told senior staff that their use of internal AI tools will be measured and used to decide bonuses and promotions. This seems like a good sign that Accenture has not deployed many AI tools that are useful for senior execs (I wonder - are they allowed to use AI to write staff reviews?). LINK Sequoia invested $1bn at a $4bn valuation in 'Ineffable Intelligence', a new 'build something about AI' startup from David Silver, who was an early employee at DeepMind. Notable partly because it's in the UK, not the Valley. LINK And Fei-Fei Li, one of the OGs of machine learning, raised $1bn for her lab company. LINK Apparently, OpenAI is working on several different devices with Jony Ive, including a $2-300 smart speaker/screen/camera home hub. LINK US military versus Anthropic Trump's defence secretary has threatened to list Anthropic as a 'supply chain risk', meaning no defence contractor could use it, after Anthropic apparently tried to set conditions on what its tech could be used for. This is one of those 'terrible person might perhaps have a point?' moments: no government can give a supplier a veto on policy, and no responsible military would put soldiers into the field using a system whose supplier might turn it off - as Ukraine has experienced with Starlink. LINK The first big 'social addiction' lawsuit Mark Zuckerberg testified in a California lawsuit that claims social media is addictive and harmful and companies including Meta and YouTube should be punished. Most social scientists will tell you that the evidence here is inconclusive at best, and one of the strongest indicators of a moral panic is when people say that they 'know' something 'must be true' without asking for real proof. In this environment, 'water is wet!' emails tend to get taken as proof of evil - hence people have got very excited by e-mails saying 'Meta targets teenagers' when, um, it was started by a teenager (Zuck was 19) for college kids. That said, I can't call the outcome of a jury trial, and more importantly, how long any verdict, appeals, US constitutional challenges, and products resets might take. Some people are very keen to call this the new tobacco, but it might be more accurate to think about car safety and seatbelts. LINK No more 'Autopilot' For at least a decade, people have criticised Tesla for calling its assistance cruise-control 'autopilot' and 'full self-driving' when it cannot drive the car by itself, and this lulls people into a false sense of security, taking their eyes off the road. Now, California has agreed, forcing Tesla to stop using the term. LINK Trump tech A Netflix board member criticised Donald Trump, and Donald Trump said that Netflix should fire them. Recall that the US government could try to block Netflix's takeover of Warner. One of Ronald Reagan's favourite jokes was that the Soviet Union and the USA both have freedom of speech: in America, anyone can say 'Reagan is an idiot!' and nothing will happen to them, and in the USSR it was exactly the same - anyone can say 'Reagan is an idiot!'… Clearly, Trump doesn't see things that way. Also, see the next item. LINK Reuters reports that the US State Department plans a website to mirror content that is banned in Europe and elsewhere. This must have sounded like a great way to get points with Trump, except that the actual content being banned (sorry, 'censored') is child abuse material, deep-fake nudes, Holocaust denial and straightforward Nazis. This probably isn't a good look even for Trump, and anyway, why bother when you can just use Twitter? LINK | 
No comments:
Post a Comment