ehsanu1

Check this site for the first part? Maybe they can offer a notifications API, hah.

For my schemas, I found LLMs really wanted to just use markdown embedded in the strings, so I've been considering doing away with the schema. I also figure that embedding markdown in a string may make it perform worse as it has to juggle nested formats, and thus escaping and such (wanted: eval for this). By replacing the json tool call with basic markdown extraction, I'd lose some structured data but gain flexibility (html would be even more flexible).

Wondering if you are referring to adherence to required data in a schema when you say LLMs do better with json vs markdown, or something else? Or perhaps to tool calls and/or strict json output being more reliably formatted for clean extraction?

That a personal website? Prod means different things in different contexts. Even then, I'd be a bit worried about prompt injection unless you control your context closely (no web access etc).

I have. In fact, I've been building my own coding agent for 2 years at this point (i.e. before claude code existed). So it's fair to say I get the point you're making and have said all the same stuff to others. But this experience has taught me that LLMs, in their current form, will always have gaps: it's in the nature of the tech. Every time a new model comes out, even the latest opus versions, while they are always better, I always eventually find their limits when pushing them hard enough and enough times to see these failure modes. Anything sufficiently out of distribution will lead to more or less nonsensical results.

That shows it knew this bit of satire more than anything. Also, the problem as stated isn't actually constrained enough to be unsolvable: https://youtu.be/B7MIJP90biM

Business problems are essentially neverending. And humans have a broader type of intelligence that LLMs lack but are needed to solve many novel problems. I wouldn't worry.

An honest mistake.

What are the numbers? Are there problems other than context usage you refer to?

They aren't necessarily "stored" but they are part of the response content. They are referred to as reasoning or thinking blocks. The big 3 model makers all have this in their APIs, typically in an encrypted form.

Reconstruction of reasoning from scratch can happen in some legacy APIs like the OpenAI chat completions API, which doesn't support passing reasoning blocks around. They specifically recommend folks to use their newer esponses API to improve both accuracy and latency (reusing existing reasoning).

But those aren't things you can really separate for proprietary models. Keeping inference running also requires staff, not just for the R&D.

I wonder how this compares to running sqlite off of an s3-backed ZeroFS https://github.com/Barre/ZeroFS

To spell it out for myself and others: approaching equivalent calculations for each individual attention block means we also approach equivalent performance for the combination of them. And with an error bar approaching floating point accuracy, the performance should be practically identical to regular attention. Elementwise errors of this magnitude can't lead to any noteworthy changes in the overall result, especially given how robust LLM networks seem to be to small deviations.

Could you substantiate that? That take into account training and staffing costs?

https://www.anthropic.com/news/anthropic-and-the-department-...

You don't get a simple request/response paradigm with claude code: 1 message from the user results in a loop that usually invokes many inner LLM requests, among other business logic, resulting in some user-visible output and a bunch of less visible stuff (filesystem changes, etc). You control an input to the outer loop: you can only do some limited stuff with hooks to control what happens within the loop. But there's a lot happening inside that loop that you have no say over.

A simple example: can you arbitrarily manipulate the historical context of a given request to the LLM? It's useful to do that sometimes. Another one: can you create a programmatic flow that tries 3 different LLM requests, then uses an LLM judge to contrast and combine into a best final answer? Sure, you could write a prompt that says do that, but that won't yield equivalent results.

These are just examples, the point is you don't get fine control.

I've gotten great results applying it to file paths + signatures. Even better if you also fuse those results with BM25.

Embedded usearch vector database. https://github.com/unum-cloud/USearch

They have rate limits for this purpose. Many folks run claude code instances in parallel, which has roughly the same characteristics.

You can't control it to the level of individual LLM requests and orchestration of those. And that is very valuable, practically required, to build a tool like this. Otherwise, you just have a wrapper over another big program and can barely do anything interesting/useful to make it actually work better.

You can: https://claude.ai/settings/data-privacy-controls

True. You can turn it off though: https://claude.ai/settings/data-privacy-controls

It has to be suited for human consumption too though.

I wonder if this has any real benefits over just doing very simple html wireframing with highly constrained css, which is readily renderable for human consumption. I guess pure text makes it easier to ignore many stylistic factors as they are harder to represent if not impossible. But I'm sure that LLMs have a lot more training data on html/css, and I'd expect them to easily follow instructions to produce html/css for a mockup/wireframe.

It kind of makes sense if you relate it to ASCII art, which is very often not ASCII for similar reasons. The naming evokes that concept for me at least. Naming is hard in general, I'm sure they tried to find a name that they thought worked best.

I agree that "TUI" is a better fit though. But not TUI-driven-development, more like TUI-driven-design, followed by using the textual design as a spec (i.e. spec-driven development) to drive GUI implementation via coding agents.

But I doubt you can opt in to them training on that data coming in via OpenCode.

I understand them not wanting to allow non-coding agents to use the subscription, but why specifically block another coding agent? Is the value Anthropic gets from users specifically using claude code that high? Is it about the training data opt-ins?

What exactly do you mean by custom tools here? Just cli tools accessible to the agent?

Doing something merely requires I/O. Brains wouldn't be doing much without that. A sufficiently accurate simulation of a fundamentally computational process is really just the same process.

The DB specifically, or the concept of event sourcing? Event sourcing is not a new approach and has a lot of similarities with temporal's approach, though temporal events are not necessarily business events and deterministic event replay is required with temporal. In the general case of event sourcing, arbitrary processing might be done on the event stream to produce some final state or do whatever needs to happen for your use case. As long as you're persisting the events and using events as the basis for your business logic and state, you're doing event sourcing.

I dont know anything about this specific DB though, if that was what you were wondering about, that's more of an implementation-level detail. Temporal server just uses regular mysql and supports mutiple storage backends.

This seems like a good template to generate synthetic data, with positive/negative examples, allowing an embedding model to be aligned more semantically to underlying concepts.

Anyways, I'd hope reranking models do better, have you tried those?

Do you assign different responsibilities to different LSP servers when there multiple I suppose?