|
Microsoft deploying OpenAI
:- As reported last week, Microsoft is making a big investment (the reporting was $10bn but with lots of ‘structure’ - sign of the times). In parallel, it’s announced it will make ChatGPT an API service available on Azure, so developers building on Azure can use it. (OpenAI itself and hence ChatGPT runs on Azure’s cloud.) There is a scramble to grab developer mindshare and create the developer and ops stack around taking generative ML to market - see also Google’s post below. But it’s also interesting to contrast this with the apparent scaling back of Hololens - that was the last ‘this is the future!’ tech that Microsoft talked up as something that would return it to leadership. AZURE, OPENAI
Apple VR/AR
:- The rumours that Apple is getting ready to launch VR this year keep getting louder (though we’ve heard them every year for at last 5 years), but meanwhile Apple has, like Microsoft, apparently decided it’s not close to shipping AR. VR is an engineering problem (make the screen better and the graphics faster) and a content problem, but AR today remains a physics problem: no-one (publicly) has optics that can give a wide field of view and good visibility in full daytime lighting (let alone deliver things like depth of field).
Training data copyright ”
:- Both ChatGPT and the new image generation models are trained with vast amounts of data mined from the public internet. But is that fair use, and if you make something using that data (and, for the imagery, the manually created metadata around it), then who owns it? Getty is suing Stability AI for using its data without permission
Google: “Hey, we do AI too! ”
:- Given the way Generative ML has taken over almost every conversation in tech in the last couple of months, Alphabet and Meta can be forgiven for wanting to point out that they’ve been working on this too - indeed both companies had GenML chatbots in testing earlier last year (and the difference is as much about approaches to risk and safety as the tech itself). Now Alphabet’s AI research group has a big blog post listing all the stuff they’ve been working on.
The privacy swamp
:- Cross-site cookies are ‘not private’ and therefore going away, and the ad industry has been trying to come up with a way to do ads that are both relevant to your interests and private. That should be possible in theory, but no-one can agree on how: Google’s latest iteration of its browser-based targeting, Topics, has been shot down. Meanwhile I keep wondering if Apple will launch an iPhone on-device ad network.
RIP Twitter clients
:- There were third-party Twitter apps long before it had any of its own, and indeed those apps invented things like threads and the word ‘tweet’ itself, defining the experience when no-one knew what it should be. A decade or so ago it heavily restricted what apps could do, making a decision to be (like Instagram) a system rather than (like email) a protocol. However, several dozen third-party apps continued, at a small scale, mostly used by users who didn’t want to put up with the mediocre apps Twitter itself made.
That ended last week. The apps all suddenly stopped working, and no-one knew if it was a bug or a decision. A week later Twitter said that it was ‘enforcing long-standing rules’ - but those rules didn’t ban the apps. Then it changed the API terms on its website so that now they did ban apps.
This is a very first-world problem, but if Elon really wants to make Twitter a ‘super app’ he probably shouldn’t be burning developers without notice. It’s also sadly reflective of the way he’s replaced one kind of dysfunction with another.
|
