Microsoft's Relationship with OpenAI Is Not Looking Good

Microsoft, for all their faults, seems to have honed in on this issue explicitly. Enterprise Data Protection seems like its designed to massage this fear, whether it works or not who knows.

I feel like you pulled those numbers out of thin air. $30 is something like 30 minutes of work or less in a typical office. Can a super-google that covers most of what you need (but not everything in every silo) save you 30 minutes per month? Yes for sure.

Well, this is what people are missing IMO. Yeah, general purpose chatbots can’t solve advanced problems. However, models and pipelines (using similar technology as chabots) specifically built for certain tasks can achieve really incredible results — like in the area of protein folding (https://youtu.be/P_fHJIYENdI)

I think there is a lot of potential, but most companies are just slapping a chatbot (which uses some other company’s models) on their app and calling it a day. Very few are taking a deep look at how ML could actually solve real problems.

Are you sure? Doesn't the tool just have to save the user less than an hour a month for it to break even?

IT Sec has these things blocked for this reason. Probably made easier by only allowing company devices on the company network.

You still have a huge build-up and a bust? Earlier real products, from railroads to the Internet, have followed this pattern.

Even more, AI has a plausible-sounding argument for being a 500b product or even $infinity dollars. I don't think it's correct with current tech but it's wrongness rests on subtle things. People who seem at least as smart as Gary Marcus (AI critic) seem honestly convinced this AI generation of is going to create AGI. An investor who doesn't know who's right is going to want to cover their bets.

Huang expressed the truth. The reasoning feature is a loop, which means more hits to the LLM, which means more compute. So more datacenters.

As much as the markets love to talk about investment values, and valuations, this really means very little. A lot of the investment is in kind, not cash, and the marginal cost of the kind is often a tiny fraction of the marginal cost.

Bailouts to US auto manufacturers occurred because they were both unprofitable and loaded with debt they could not pay for. Ditto the bank bailouts of 2008.

MS paid billions for Skype and billions more for Nokia. Nokia (windows phone) died soon, Skype will be turned off this May. Neither caused a stir, when your cash on hand is measured in tend of billions big investments are not risky.

Because bullish articles are bought by AI companies and/or encouraged by editors because AI is the hot traffic keyword right now, hence tons and tons of articles about utterly banal ML integrations being rebranded as AI so Wall Street will think your fuckin Microwave company is ushering in Skynet.

And on the flip side the bearish articles are because we're now almost 2 years into this hype cycle that Silicon Valley is trying so incredibly hard to make us believe is real, and we've had, yes, a few minor interesting things, but the vast majority of every part of it is unwanted integration of chat-bots into products that do not need chat-bots, or the replacement of lower-paid staff with a ChatGPT integration that doesn't work.

There is, and I cannot stress this enough, so incredibly little consumer interest in this tech. Nobody. Fucking. Cares. In fact, I read something recently that including AI in product descriptions actually hurts the rate of purchase by people because they're so fucking sick of it.

Now mind you, that's not going to kill AI companies, because AI companies are not worth 50 bazillion dollars on the back of chatbots. They are worth all that because of the "usefulness" of varying legitimacy in the defense industry, the insurance industry, border security, all manner of things that will almost without fail make your or other people's lives meaningfully worse, but they'll do it in a cost-effective way that reinforces existing capital and power. That's why it's worth so much money.

There is no doubt that they will get one if they need it so why worry about it?

Your subscription includes not just the stellar search engine, but access to their assistant as well, along with all their privacy benefits

What I love about assistant specifically is that it lets you switch between one of 10+ models. If you are curious for example how a different model would answer a prompt, within the same thread you can change the model and re-ask your question

Gemini, deepseek, Claude, mixtral, ChatGPT, etc

Definitely worth checking out - has definitely help me take advantage of these bots more

Due to sneaky website scraping by OpenAI many sites now have their content locked behind login, have their APIs severely restricted and watchout for scraping.

Unless you want to build a specialized AI built on specific open or closed training corpus, huge training data is required to build something similar to OpenAI's models.

Google with it's crawlers and associated relaxations should not be affected much by those restrictions.

Miscrosoft can easily update its browser/OS ToS and say "we'll use all your browsing data for training, anonymously ofcourse" and there's not much users will do except shrug.

Is there anyone else who is better positioned to collect the content for training or am I placing too much importance on training data?

This model has proven to work in places like retail. Literally how Amazon got big.

I think they are trying to use this strategy for AI as well, but at least right now there is so much competition and investment that it's tough to pull off.