causal

I sometimes suspect AWS deliberately looks for ways to extract low-overhead tasks into dedicated services for the simple reason that many people will pay for the service without thinking about whether they really need it.

That makes sense, and if that's the reason it's another vote for open models

Anything not JSON

Shocked this wasn't already a feature. Bummed they only seem to have JSON Schema and not something more flexible like BNF grammar's, like llama.cpp has for a long time: https://github.com/ggml-org/llama.cpp/blob/master/grammars/R...

I reaaaaally wish we could provide an EBNF grammar like llama.cpp. JSON Schema has much fewer use cases for me.

If it's not profitable, it's running on capital. Subsidized.

> It's not actually subsidized

Source?

Wouldn't be the first time an Italian prosecutor went on a witch hunt based on wild conspiracy theories: https://www.cbsnews.com/news/monster-of-florence-amanda-knox...

I wonder if a price correction would be a boon for open source, with the economics of smaller / self hosted models making a lot more sense when API prices have to surge.

Yeah the "spiral bar chart" was an artistic but maybe not practical choice.

Yeah there has been some lamenting at all the money being thrown at technology hasn't been for anything truly game changing, basically just variations of full stack apps. A few failed mooonshots might be more interesting at least.

"Study finds..." feels clickbaity to me whenever the study is just "we found some randos on social media doing a thing". With little effort a study could find just about any type of person you want on the Internet.

Yes that's part of what I meant by read, but understanding a codebase goes a lot faster when you begin with some higher level understanding.

IMO this is why Claude Sonnet is better than ChatGPT: Sonnet is so much better at clarifying, drawing diagrams, writing documentation. It TRIES really hard to keep you in the loop, but of course you can choose to ignore everything it writes and just say "do more" without understanding anything.

What did you expect, complete arpeggios and site-reading Mozart? In six months he taught a wild animal to go from nothing to understanding how to interact with a complex instrument, and do so with increasing sequences in response to a reward. That's really cool. Your expectations might be a tad high.

I was confused at first too - the story is in sections accessed at the top

Yeah I believe this is basically Taleb's investment strategy (or at least, what he seemed to claim in his books)

That's sort of Taleb's whole thing: most successful investors just happened to be riding a wave and cannot claim any sort of genius. The Black Swan event is always explained away as some kind of environmental blip and investment success explained as the inevitable outcome of hard work- when in fact it is Black Swans that are inevitable and hard-working investors are just a background constant.

Hopefully as a society we can also learn the lesson that tech companies cannot be trusted to deliver what's best for us.

It makes me so sad that it's possible for technology to steal the need to talk and play, even from our youth. If you have little kids you know how frantically they NEED to yap and play. I hold such horror for anything that would sap such life away.

Google's consumer billing is a nightmare to manage and I've been fraudulently charged (in my opinion) without recourse at least once.

Neato. Sensible updates, no loss of support for previous versions? The software dream.

Secret to Azure success: Just wait for the others to f it up.

Sure let's just remove all security, encryption, firewalls and auth - nobody will abuse vulnerabilities if it's a crime!

Totally. I think the original "Lethal trifecta" post by OP only pertained to data exfiltration and never included changing state (maybe was implied by sensitive data access).

Rule of 2 model has holes.

I think the rule of 2 would work if it kept the 3 from your lethal trifecta. "Change state" should be not be paired with "communicate externally".

And even then that's just to avoid data exfiltration- if you can't communicate externally but can change state, damage can still be done.

But you're still generating code to be rendered in the browser. Google is a few steps ahead of this: https://deepmind.google/discover/blog/genie-2-a-large-scale-...

What is meant by "residual connections" here?

But I bet you could reconstruct a plausible set of distributions by just rerunning the autoregression on a given text with the same model. You won't invert the exact prompt but it could give you a useful approximation.

Yep. People love to make blanket statements like this without remembering how hard it was for parents to supervise Internet use even when there was only one Desktop computer in the family area. Nowadays you can almost any device to lookup porn.