kooi

Local model utopia!

Speaking annecdotally because im an engineer, not in sales:

Our company has secured projects and executed them to a standard which we would not have otherwise were it not for AI or hiring more people. Now, these are at least 3 year projects, so we are only seeing ~3 being AI accelerated.

For us, yes, it's a multiplier, directly translated to more $$.

Not sure what all these words mean. These are mathematical results, I'm not sure why people are acting like they are anything otherwise.

Mass media doesent do ML reseaech.

The results are in the math. Does not really matter the terms used.

You wont get anything materially useful, apart from an intuition, from using different terminology.

> 33M-parameter text JEPA reaches .125 on the PRH paper’s benchmark, against .034 on Flickr captions, level with Qwen2.5-3B and -7B. They need fewer samples.

Im bullish on jepa models. Success of autoregressive next token prediction feels like a stroke of luck.

Not a security person, but are there standards for PII encrypted at rest?

Surely this is common? Surely?

As long as you make sure the end points are valid, this is most intuitive for me. Just using i as a backwards step.

for (int i = 1; i <= N; i++) {

    array[N-i]

}

Thanks for the reference.

From the abstract:

The alcohol ethoxylates present in the rinse aid were identified as the culprit component causing the epithelial inflammation and barrier damage.

Its outrageous.

Thing which are public law should be publicly accessible.

There's an online portal for NEC code book, but they made it awful, no search, no hyper links, the bare minimum.

I used codex in computer mode to collect the relevant pages to my project. (Electricity to shed).

Looks really neat. I'm keen to try this on my clicks communicator when that arrives.

It's mediatek NPU claims to handle ONNX. That plus the 12GB ram makes the wonder if I can get this phone to be a mini-agent.

It's a call back to comments complaining about math notation: https://news.ycombinator.com/item?id=50122878#50126010

Watch out, this paper has scary ~math notation~

;)

Coming from Rust:

Why is does implementation still have a the generic T? Should it not be concrete on i32?

/// i32 forms a sum reducer under 0 and addition const sum_trait = impl(Reducer(i32), struct { pub const empty = 0; pub fn combine(x: T, y: T) T { return x + y; } });

Wavelength = 1/f

Yep, the context switching let's me do 2 or 3 active agents, but I have to slow down if real issues come up.

Maybe the agent maxing dudes are doing their 12+ agents on a monster vibe project and all they say is "fix it", "make it work" all day

Design Patterns by Gamma et al.

Recently helped me solve a non-obvious code-stink design issue.

Gezz so much vitriol.

> Freeze frame photos show what appears to be a lightweight hollowed out car shell prop, likely using magnets, suction, and compressed gas to help the country's richest nazi once again lie to the gullible American public and boost the valuation of a car company that stopped being innovative more than a decade ago.

It's obviously a stunt. But IMO it's a real stunt probably developed by the controls engineers at SpaceX to show off some pretty crazy control capabilities.

This is NASA SkyCrane on steroids.

Gotta pay if i make more than 20 mil a month?

Sounds like a great problem to have.

My question: What kind of dynamical system observability does Timestral provide? Is it a blackbox y = F(x, params) model for offline use? Can it provide the actual dynamical system equations? Or are you stuck forever querying the Timestral server (which is bad, think about if you're grid system operator doing day ahead, month ahead forecasts, and Timestral goes down)

Typically, in linear, non-linear, and NN models, we fit some noisy data to A) An underlying model with unknown parameters or B) A model free search which attempts to construct combination of "basis" functions or C) A purely black box model which we only care about Input/Output error.

A) is useful because often we know an underlying model we want to match, but don't know the parameter values. If we can determine the parameters, we get insights about the real system. E.g, how are these two variables coupled and how strongly -> We can make business decisions about what happens if we adjust some parameters.

So what exactly does this timeseries foundation model do and provide?

Maybe, the observability is that you have a forecast, and you adjust the covariates and see what happens. But one would want much more insights than a "executive summary" of the forecast.

The important parts for me as a signals processing R&D engineer.

> Classical models such as ARIMA and linear regression remain well-established tools for forecasting time series. These models are usually fitted to a single series and learn only from that series' history.

> Deep learning models such as PatchTST add the capacity to learn nonlinear patterns across many related series, but they are still typically trained from scratch on each dataset, so what they learn doesn't carry over to new domains.

> Time series foundation models bring the pre-training approach to time series. Trained on a large and diverse corpus, our model learns patterns that transfer across series, forecasting zero-shot on data it has never seen, even in domains where data is scarce.

Edit: Looks like it's a pure black box which you can run on premise.

> Output: Probabilistic forecasts, 21 quantiles

> Inputs: Multivariate targets, past-only and known-future covariates

Goodness, someone posts a great article which IMO accessible math, and people get upset.

If you think that is crazy, ohh boy, just wait til you see all the other papers underlying LLMs, AI, and optimization in general.

I appreciate this post because starting from linear algebra is the correct way to form an intuition of the higher dimensional semantic spaces we're dealing with.

Very cool.

I'm excited for my Clicks Communicator to come in. I'd love to try to connect this to a hot-key for easy access.

My gmail got absolutly flooded with LLM generated spam messages. They all had "Hello, test" in many different languages, correspondence personas, etc.

which makes me think someone was trying to test their mass email spam generator.

1000+ emails bypassed my spam folder, went straight into my Primary inbox. From the POV of the "tester", I'd say they had a success.

I'm thinking of migrating off Google for proton mail now.

Yes, it's crazy. Just keep in mind that 80% of the economy running outside of our little software bubble is still using bulldozers, shovels, paint.

"It’s 2026 and it seems like AI has reached singularity." No, no it has not. The singularity, mathemetically and as originally termed by futurists, is when progress reaches an asymptote and by definition becomes undefined.

Just had a look. The OP library claims to be no_std compatible while ChoRus is not.

The 1%ers are in the vertical, but the question is vertical to where?

It needs to a potential field with practical, economical, "real life" attraction well. I.e, robotics, real economic efficiency gains, manufacturing novelties.

The worry is that the 1% is attracted towards a non-practical money hole. I.e: Token burn for the lols, sophisticated software systems that dont provide actual value outside of giving NVIDIA cash.

It's rough, but at some point there needs to be an arbitrary line.

I think 21 is used is many cases because that's supposedly when the brain has completed most of it's growing or something about pre-frontal cortex being fully developed.

The crisis in reading is due partly to trendy, faulty, and "inclusive" pedagogy.

A return to classic phonics will help a lot. This is already happening because the empirical evidence against the whole language approach is too sound to ignore.

See this study from 2013 https://eric.ed.gov/?id=ED545621

Not quite sure what all those words mean.

Dynamical systems will never be solved in a semantic domain.

IMO, they can be helpful the in robotics stack, from planning level up, but that's it.

I think where both camps get hung up is sometimes the process method group "ignores" the obvious outcomes and effectiveness of LLMs.

But the outcomes group "ignores" the fundamental limitations of models which are purely text based.

E.g, a baseball players trains to catch high-speed balls and they dont do it by: "ball velocity 50mph, vector:[1,2,3], run move hand command now"

That's absurd.

No, there is an embodied network which is "trained" on visual, tactile input, and control as direct output.

LLMs are fundamentally not the right tool for that.