Figma bounces back Fresh from the cancelled $20bn Adobe acquisition, Figma launched its developer conference with big splash around generative AI. The Adobe deal was premised on Figma being a cloud-enabled disruptor of PC design software, and just six months later it's looking over its own shoulder at LLMs, with the more excitable 'AI maximalists' wondering if Anthropic or ChatGPT can just… make you a layout. Of course, the reality is more complex, and it's better to think in terms of automation and autocomplete than vertical software and replacing designers, but whole layers of grunt work will be removed, just as they were with GUIs and SQL. LINK AI product Anthropic has started building out team collaboration tools. One big question for the next couple of years might be whether the general LLMs become useful for specific, vertical tasks faster than the vertical tools get LLM features. LINK Meanwhile, after hiring Kevin Weil (Twitter, Instagram) to run product, OpenAI bought two analysts startups, Multi and Rocket, also focusing on building out a useful product set. MULTI, ROCKSET Music labels versus AI music generators There was a joke 20 years ago that instead of investing in music startups, it would save time if VCs just wired the money directly to the record labels (and their lawyers). Now the labels are suing several of the buzzy new AI music generators, while YouTube is trying to do deals. Suing everyone didn't save music sales, though, because the toothpaste was out of the tube, and it seems likely that the same applies here: the models will be public and enough of the training data will be too, so millions of people will be able to make 'music that sounds like Taylor Swift' on their PC. I wrote about the problem last year (see above). LABELS, YOUTUBE The week in AI Google published its latest 'Gemma' open source model, bouncing back to the top of the open LLM benchmark tables. LINK Character.ai doesn't get much attention, but it's quietly one of the early breakout consumer hits of generative AI, with people doing huge volumes of chats with AI-generated avatars. LINK NBC will use generative AI versions of its commentators to provide scaled, automated material for its Olympics coverage this summer. LINK OpenAI is delaying the voice assistant (together with the 'Her' controversy) it announced earlier in the summer to work on safety concerns. LINK I'm surprised it took this long, but OpenAI cut off API access in China. Good news for Chinese AI companies, which are not far behind - except of course for their access to Nvidia chips. LINK Google used AI to add 110 new languages to Google translate. Age of Wonders. LINK And, a startup called Etched launched a dedicated ASIC for transformers (the maths model inside LLMs) to take on Nvidia. I am not a semis analyst, but the track record of new chip companies isn't great - cf Graphcore's struggles (indeed, Nvidia itself is one of of the rare exceptions). But if it works... LINK, GRAPHCORE EU versus Microsoft Very predictably, the EU has charged Microsoft with breaking competition law in bundling Teams with Office. Two challenges: first, yes, this does make life harder for Slack etc, but, if we took this logic at face value then your word processor would not be allowed to include a spell checker and your spreadsheet could not do charts (in the 1980s both of these were separate purchases). What's the coherent theory for how we choose the trade-off between integration and competition? Or are software companies just not allowed to add features once they pass a given level of market share? And second, whatever your view on that, it took the EU far too long to get to this: it's too late. LINK Amazon, Temu and Shein Amazon plans to launch its own marketplace to aggregate Chinese manufacturers, taking on Temu (though one of the claimed reasons for Temu's success is that many of the manufacturers were pushed of Amazon Marketplace previously). Meanwhile, Shein has filed confidentially for IPO in London, which means we will soon get to see some numbers. The last reports were that Shein did $45bn of GMV last year, which would make it the largest apparel retailer on earth. AMAZON, SHEIN Software eats cars? VW stumbled badly trying to write its own car software as it transitions to EVs, and now it's done a $5bn JV with Rivian. Making cars at scale is really hard (hence all the EV bankruptcies in the last few months, and the hard-scrabble bootstrapping at Tesla), but so is software. LINK Remember the Rabbit Phone? Rabbit got a huge amount of attention in tech for a brief moment this year, based on a demo that looked magic but made little sense and was probably too good to be true (plus a cool hardware design from Teenaged Engineering). It's all been downhill since then, and this week someone worked out that it embedded all the API keys (OpenAI etc) in the device, giving anyone who cares full access to all user data. Oops. LINK | 
No comments:
Post a Comment