gradys

It can be the case that both:

- The physics of the universe can be completely modeled as computation, and

- It's possible to pose undecidable problems about the way the universe unfolds

This is intrinsic to the idea of undecidability even for Turing machines, e.g. "we equate computation with the functioning of Turing machines, but there are real processes executable in Turing machines that are undecidable".

IPOs pushing up housing prices

New Waymo car models rolling out

The laser thing in front of city hall

Pride month, parades this weekend

Antisocial guy harassing Scott Weiner on social media at Dolores Park

Lurie administration, after initially making progress on crime and cleaning up the streets seems to be stalling out

Supervisor Jackie Fielder back from long mostly unexplained absence that was maybe due to her illegally leaking confidential information

This was long before. Google had conversational LLMs before ChatGPT (though they weren’t as good in my recollection), and they declined to productize. There was a sense at the time that you couldn’t productize anything with truly open ended content generation because you couldn’t guarantee it wouldn’t say something problematic.

See Facebook’s Galactica project for an example of what Google was afraid would happen: https://www.technologyreview.com/2022/11/18/1063487/meta-lar...

Both would be understood and are roughly interchangeable.

"Sequential" feels more appropriate to me for the task runner scenario where we wait for one task to finish before running the next.

"Series" suggests a kind of concurrency to me because of the electrical circuit context, where the outputs of one are flowing into the next, but both are running concurrently. Processes that are Unix piped into each other would be another thing that feels more like a "series" than a "sequence".

You might be using JSON mode, which doesn’t guarantee a schema will be followed, or structured outputs not in strict mode. It is possible to get the property that the response is either a valid instance of the schema or an error (eg for refusal)

Is it luck, or is scaling arithmetic genuinely a useful capability to offer the world?

I assume you know how to program in Python? I would start with just the client libraries of the model providers you want to use. LLMs are conceptually simple when treated as black boxes. String in, string out. You don't necessarily need a framework.

I’m not as familiar with the Baby Bells, so this is a surprising comparison to me. Bell Labs was famously so productive while it had the monopoly money hose, and not as much came from it after Bell was broken up.

What are the most noteworthy accomplishments of the Baby Bells?

Sounds more like Generation Augmented Retrieval in that case.

I don’t know of any public details on how they implement Context Caching, but that is presumably exactly what they are doing. Just caching the text would be a minimal savings.

The size of the cached internal state of the network processing the book is much larger than the size of the book. The resource that is preserved with caching is the compute required to recreate that state.

I’m guessing there was an instrumental reason for this, for instance to check that the model was listening before launching into what they wanted to demo

There are many different rules of thumb on this floating around. 1g per lb of body weight is indeed something a number of well informed people recommend, but those people generally acknowledge that it’s something like an upper bound on the amount that is useful.

Cardio isn’t super important for what? It certainly has longevity benefits over and above those from resistance training.

Attention itself was the key idea of that paper and, as you sort of acknowledge, was definitely not just throwing things at the wall. It was the culmination of a long line of work gradually progressing toward fully dynamic routing via attention, and it was motivated, if not by deep theory, at least deep intuition from linguistics. The other details of transformers are perhaps sort of arbitrary, but made sense to everyone at the time. There was no claim that those other details were optimal - just that they were one way of surrounding the attention mechanism with computing machinery that worked.

The key feature of the NEC is that the cities are large and close together. This is true almost nowhere else in the country.

What does understanding and reasoning mean?

I'd love to be able to get a print of the map with a customized zoom level and origin point.

This is awesome! I'd love to be able to bookmark this with a specific starting location set in the URL.

https://grady.io

You can achieve this without UUID keys if you attach UUIDs to the requests. This has the advantage of working for all kinds of mutating requests, not just creates, but requires that you store these request UUIDs for some period of time as well.

My intuition on this:

Maximum likelihood training -> faithfully represent training data

Reinforcement learning -> seek out the most preferred answer you can

It might be legal from Canada's perspective, but that doesn't necessarily mean legal action couldn't be taken in the US. Is there something specific to the law on recording phone calls that makes this not a problem?

“Symbolic reasoning” used to make “decisions which influence the environment with which the system interacts” could describe almost all real world computer systems, could it not?

If everyone chooses to comply with the law in the same annoying way, that’s a consequence of the law.

It may be a second order effect, but good governance means considering second order effects.

It doesn’t seem like this article considers any benefits from allowing this launch or specifies a threshold of damage under which the author would be okay with allowing the launch.

In general, doing things like developing the capacity to launch very large payloads to space will come with costs. We wouldn’t be able to do any such things if we weren’t willing to pay those costs to get the benefits.

We don’t know if any spies are going to get burned or if any methods will be discovered and shut down. There was intelligence on specific things individual foreign leaders said in private. Combined with other info, that could definitely lead to spies getting killed. This is often the way spies are discovered.

The model is indeed so overpowered that it doesn’t matter in practice. See the Sentencepiece paper for some discussion of the design decisions on stuff like whitespace.

This sounds like a long running conversation. Are there problems with extending past the context window?

One could imagine an alternative where the API manifest was provided privately to OpenAI in the developer console or where the plugin developer had to implement an OpenAI-specific API structure. Doing it this way is more, dare I say, open than it might have been.