nostrebored

The idea is that you can describe the predicates, producers, and goals. If I’m understanding correctly. This is how a ton of stuff works: excel, your IaC, etc.

Here the goal can be a render state and it is the solvers problem to figure out what should be there.

So you define these things very atomically, but the behavior automagically finds a valid render state (usually through converting things into a standard planner or solver language, haven’t dived into this repo yet)

We will continue to reinvent dataflow and planning problems until the end of time; you will be surprised how many fundamental pieces of software use this paradigm

But how do you verify a fair or representative distribution of complexity?

I don’t think it is revelation, it feels like active incompetence. You can lay out every decision and watch Claude/OpenAI actively fuck up in inexplicable ways.

If you can’t tell value at a 25 (or 10 or 5) person org, you will not be able to at a larger one.

If you weren’t making intentional decisions about cloud vs on prem… why? I worked with many small businesses that did.

This feels like a value-divorced outsourced engineering mindset. You have the power to understand if things are good ideas or working!

I don’t think it has much to do with technical jargon. You are making a product for technical people, jargon is fine.

You should not write things with AI that you expect people to read

I have it on $200 Ultra and it feels like Opus 5.5 (I do not think about limits) rather than gpt-6* models where I constantly hit a wall.

yes, I’ve only had it really degrade once and then I used a different agy session to fix the session table. I’ve been using maybe 5-6 concurrent sessions and it’s largely fine

Claude can usually make a cli that uses the endpoint and the keychain secrets now. Eg i have ‘cluse’ for muse in cc.

Why would you think so? From the info here it could be measurably better?

I’ve been primarily using agy cli and argon for the last week and I’m fairly satisfied. Rarely am I looking to go back to opus and it seems much much better at model training and conceptual math

Was there any intent behind the audio encoder representation or was it frozen backbone -> co adaptation?

Is it a parameter count problem? Ie why are you not seeing the cross modality bump in performance?

They make very good classifiers and are easy to fine tune for specific use cases.

You can train gradient boosted trees on jev style features, it is a quick “good enough” approach for a lot of stuff

right, and this would be a positive for the world.

they are, I believe the third clip is dystopian AI girlfriend

I strongly like muse-wonder-full and it's broadly competent

LLM systems start to Frankenstein things together at a very modest amount of code. The lack of abstractions cause huge and glaring problems for anything that needs to run predictably.

Abstraction is an issue for every AI coding agent. They are painfully useful, but need constant badgering in the right direction.

There has never been a better time to write frameworks that constrain how things have been done.

Almost all of our work is on our framework internally. The latency, guarantees, build time errors, and feedback that can enable vibe coding individual implementations.

I find myself constantly wishing for more framework code; I wish we had built our original framework in something that threw angrily when we abused it more.

I wish our type system were more expressive, I wish our threading story were better, I wish I wish I wish.

It’s never been easier to rework the whole thing, and I still am happiest when our framework is artisanal.

I literally run an AI startup; we all talk about the clankers and have for more than a year

lol what are you talking about, clanker is a pretty ubiquitous term in sf right now

But why would you have to make the succinct suite by hand? Not what I was suggesting, I have gotten in the habit of sending extremely concise, human readable repro scripts made by Claude.

Partially because I can read and validate it actually shows what it claims to show, and partially because I expect people want to know what I’m telling them.

Using everything modern and useful!

Performance testing has always had problems with isolation, mocking, covariance of services etc. I’ve just spent two weeks driving down latency across our framework, and autoresearch was definitely not a viable path. Most of these loops have these logarithmic, non-step change curves.

Just like with actual research, the real benefits came from me asking why is it like X and not like Y?

The original performance testing framework Claude created to bench our different versions against did not even mock high variance provider calls…

Brother are you okay?

Would you define two apartment units next to each other, with the same floor plan as fungible?

What about the house next door?

For the purposes that people use them for, there is an immense allocation problem in the US. People are sitting on a 5b/5ba and living alone. Bizarre sets of incentives have culminated to make this possible. We have collectively decided that for some reason, it’s completely sane to live in a house that you don’t need. We should look at it the same way as owning a stretch hummer.

I realize this is probably too old to see; try without openrouter. Openrouter has caused all sorts of degradations and exotic bugs for us (not just on decision models)

There is no inalienable right to live in a place because you were there first.

Housing is a fungible good. I have friends and family living in squalor because people have decided sentiment is a valid reason to keep people in these four walls instead of those four walls.

I think people’s primary residence should not be seen as assets in the way they are today at all. The only reason they are is substantial government intervention.

Right, but performance testing has always been hard and LLMs are awful at it. So maybe your coworker was right?

I’m definitely on the “just use the clanker” end of the spectrum. But knowing when and is good. Constantly having to argue with people’s reposting of Claude’s idea of what’s wrong has left me burned out. At this point, if I have any idea about what’s going on and someone tells me “Claude says”, I will ignore them.

In this case, it sounds like you could have made a ten line repro that shows the problem. Why not just send that, something human interpretable?

Jev does not have the properties you’re talking about. It’s not very good at general knowledge, it frequently approaches chance level classification on business domains, and there is so much room for someone to do better.

Doing better, I suspect, will be throwing a bigger model underneath and doing something diffusion like (we’ll allocate a whole block to compute rather than just the constrained space) that doesn’t compromise on latency.

Yeah agreed that vision is a nice fit here. I’ve gotten some really impressive results by just throwing xgboost on 25 classification related questions.

System one models do not have an obvious use in voice ai, unless you are choosing from a list of potential utterances (eg embed -> top k -> choose from top k)