jimbo_joe

It's hard to understand decisions like this from an organization offering an alternative to Github since it goes directly against the network effects that they are supposed to be trying to build. I understand the point of banning a project after it comes out that they've commited crimes, but I hardly want the host of my repo to be the authority deciding whether my project is good or bad before the fact.

You move your mouth and tongue as if you are gonna make the sounds, just don't exhale the air. They have a demonstration video in the blog post. As somebody else have already pointed out, it's a great tool for interacting with always-on camera devices like smart assistants, makes it a bit less weird.

Hey, how do you plan FIRE with children? My daughter turned one recently and we're planning to have at least one other child. While I can see myself FIRE and be able to live off savings at some point I feel that children costs (education, general upkeep like housing, food, and other stuff) x number of children for the next 20-25 years (roughly until they become self-sufficient) coupled with economic uncertainty are unpredictable enough that I can't imagine myself FIREing when I have my children to provide for.

The blog post nails the problem well: with distributed training handled by many teams the problem becomes organizational rather then tech-based. Having a technology solution to more easily persist pipeline and artifact configs is a good idea, but it only yields an incremental improvement. IMO this can only solved with a shared culture around experimentation which has to be maintained and reinvented as the company (and hence the teams responsible for training) grows. Basically, if every layer of the stack from tooling to training to applied are aiming for repeatable training with predictable capability improvements that end users can see, then it's more likely to happen.

> Search, after Bert, is very much a commodity. > I get better results than Google on segments of common craw using a desktop computer and a research model.

For data which hasn't changed since knowledge cutoff - for sure, but for real life web search, being able to get fresh data is a hard requirement.

Pre-ChatGPT OpenAI produced impressive RL results but their pivot to transformers was not guaranteed. With all internet data, infinite money, and ~800x more people, Google's internal LLMs were meh at best, probably because the innovators like Radford would constantly be snubbed by entrenched leaders (which almost happened in OpenAI).

> For example Google is in the amazing position that it's search can become a commodity that prints a modest amount of money forever as the default search engine for LLM queries, while at the same time their flagship product can be a search AI that uses those queries as citations for answers people look for.

Search is not a commodity. Search providers other than Google are only marginally used because Google is so dominant. At the same time, when LLMs companies can start providing a better solution to the actual job of finding answers to user queries, then Google's dominance is disrupted and their future business is no longer guaranteed. Maintaining Google search infra to serve as a search backbone is not big enough for Google.

Didn't work for me on the first prompting (got a 10-pointed one), but after sending [this is 10 points, make it 9] it did render a 9-pointed one too

This will immediately kill some of the parts (Chrome, Analytics, Alphabet) and increase the antitrust issues in others (Cloud, YouTube)

Stack is usually faster to use since it's always sequential in nature which is related to why it's considered finite while heap can theoretically take all of OS' available memory. This way the user doesn't have to consider maximum stack depth prior to running the app