_boffin_

That statement gives a lot of insights into the possible model size. Llama 3.1 405b runs @ ~900t/s: https://www.cerebras.ai/blog/llama-405b-inference

From initial vibe research (which is totally not correct by any means) ~13.5k concurrent streaming clients capacity

I am floored at these achievements. Such amazing work.

If I may ask, when you started thinking about achieving this, what were the first attempts, ideas on how to go about it? What were some of the obstacles that had to be overcome to achieve this ?

My question is: what will this do to Ceberas? It validates them, did they just have their lunch eaten?

Can you go into more detail about your setup and use cases?

Probably on a business / Enterprise plan, which has managed settings and also telemetry export. Give it a collector endpoint to export to and then have collector send to s3.

The thing that I keep thinking about is the accounting / charging when it downgrades automatically.

Do they adjust the price of the api request so that only the tokens that were utilized by fable get charged at that price and the remaining tokens that the cheaper / nerfed (fable) model utilizes get charged at that price?

If the answer is no, could that be construed as fraud?

The "knowledge graph" has been there for a long long time....

Why even need soft delete in this case instead of just making an archive of all data and then deleting?

This isn't related to Dynamic Workflows, but more on the telemetry / observability side of things.

Why'd you guys not want to allow the traceparent in hooks, but allowed the session.id? Any plans on changing that?

Gonna have some fun with this!

Reminds me of automating FiServ in VBA. Omfg, but still worth it.

Can you go into more depth about this

What a statement. What a statement. How many financial institutions do they support? How many different vendors supply the platform for those institutions? How many of those financal institutions (FI) don’t support oauth or other APIs? A lot! Then ask yourself: how do they talk get the data if no api? Web scraping. Then ask yourself how they build the scrapers for those? Where do those accounts come. Employees of the company who open up accounts at those FIs? What about all the other FIs? Where do you think those come from…? How do you think that process is secured? Think the process is secured enough to make you feel warm and cozy? When the scrapers are working, how do you think they get past the security measures? Do you think those financial institutions might think it’s odd that you’re logging in from multiple IPs and that one or more of those ips might be from a residential proxy network?

The result is that I attempt, at all cost to not use anything that requires plaid or their competitors since I know how that sausage is made.

Wasn’t there something where his comp package mandated 50b in market cap?

Ahh. 100b https://investor.gamestop.com/news-releases/news-details/202...

and this sums it up right here.

> Frontier models score ~90% on Python but only 3.8% on esoteric languages, exposing how current code generation relies on training data memorization rather than genuine programming reasoning.

https://news.ycombinator.com/item?id=48100433#48102985

Please distill instead of having me navigate off site. Include link for additional info.

edit: side -> site

Read the first few comments and surprised I didn’t see it, but training data. The voluminous amount of Python in the training data.

I could write in brainfuck with ai, but I presume, wouldn’t get the same results than if going with python.

My follow up question: with AI now, why care about a lang until you need to?

> Siri fell behind due to how good Apple’s privacy is.

Uhh. What the heck are you talking about? I’m calling straight bs on this unless presented rational.

Siri has access to knowledge.db or whatever it is, which is the centralized hub for pretty much all things. Siri phones home every request made via Siri.

I think you got sold snake oil

The way i look at it is: those users are going to ask differing questions than engineering that may lead to possibilities not considered, thought of, believed possible, etc.. which can be a good thing, when harnessed correctly*.

I'd love to hear more about the positive effects of designers and PMs using AI, especially more on the PM side, if you care to go into more detail

How’s this actually going? I’m sure there are issues, but is it actually fruitful?

What i don't understand about "computer use" is why they're not just grabbing the window handles and storing them to determine what should be clicked after the first few iterations of using that a specific application. if a new case / path / whatever is found, drop back to screen grabbing and bounding boxes and then figure the handles that are there and store after.

idk.. not really thought out too much, but has to be better

Can you provide some references to what you’re talking about

I'm really enjoying the way it writes and its tone.

Is your friend on the JAX team?

You state your hypnosis quite confidently. Can you tell me how taking down authentication many times is related to GPU capacity?

Wasn’t there something with 7.1A and that the canary was gone after that version too?

And next, we’ll get to tag based file systems

What was the main focus when training this model? Besides the ELO score, it's looking like the models (31B / 26B-A4) are underperforming on some of the typical benchmarks by a wide margin. Do you believe there's an issue with the tests or the results are misleading (such as comparative models benchmaxxing)?

Thank you for the release.

What issues do / did you have with HTTPx?