nojito

Correct. We were chronically understaffed and now we are able to actually breathe and can argue that we are finally right sized.

Happier engineering teams? Happier end users of the data?

What does useful mean to you?

most effective way to debase American frontier labs

You're not going to debase the frontier labs through distillation.

a massive number of data engineering agents. We are able to stand up, test, audit, deploy data pipelines much more efficiently with the foundational models.

Why arent you using /goal?

Cloud AI is 100% worth it. We've been able to do so much more with so much less.

Their quality metrics have increased steadily over the last 10 years

https://healthy.kaiserpermanente.org/northern-california/pag...

There’s no best healthcare for everyone but that doesn’t change the fact that Kaiser is objectively one of the best.

Correct. There’s no uniform healthcare solution that works for everyone.

But it doesn’t change the fact their protocols are top notch as reflected by their quality metrics.

Not sure if you’re serious but just search the name Kaiser in any research search engine.

https://pmc.ncbi.nlm.nih.gov/articles/PMC8032167/

Their secret sauce is their ability to standardize protocols throughout their entire organization.

Kaiser has some of the best healthcare outcomes in the country/world due to their protocols and how good they are in ensuring adherence to them.

It’s going to be very improbable that these statements are true.

Both parties will just settle.

Apple already caught former employees accessing the Apple internal network with unreturned laptops after termination that’s pretty much game over.

If I were to guess this is to stop distilling and all of those blackmarket resellers.

if you can't write code, you can't review it.

I don't write code anymore and I doubt I ever will ever again.

On the flipside I review exponentially more code than ever before.

Just ask Claude to dump out assembly, or a compiled binary, but no, they don't trust the LLM that much

It's not "not trusting" the llm its that the llm has been undergoing reinforcement learning is on coding. Plus generating assembly is extremely token inefficient.

They just aren't good at agentic work.

Also risking it all for some distilled models is a recipe for disaster.

bring it in-house

People don't like to hear this but the open models just aren't good for end to end agentic workflows.

There are some very very good small open models that can excel in certain finite bounded tasks, but the foundational models are essential to building out agentic pipelines that actually work.

Your tools don't render the file though and python-pptx hasn't been updated in 2 years.

And most of the time I see some basic workflows. Summarizing Slack. Answering emails. Doing scheduled scans. Performing research and booking something. Sending emails out of Claude.

This alone will improve the lives of so many people. The real issue with AI commentary is that everyone is guilty of the hedonistic treadmill. We constantly need it to do more and more to get that sense of awe.

The best way to parse pdfs is to convert them to images and feed them into the llm.

This workflow is highly optimized.

It creates perverse incentives.

It also helped vault America into being the wealthiest country in the world.

The biggest misconception that people have when modeling using tabular data is that more data = better model.

I'd argue now is the perfect time because of the growth of agentic programming.

You're over inflating the S which is expected to increase as now they are "going to market" G&A is within expectations.

Revenue is still growing faster than costs and gross margins have continued to improve.

The real question is when they can start spending less on R&D and still compete.

All models almost certainly can’t do that.

There are many other ways to determine where schools and hospitals are needed, such as aggregate enrollment and admission statistics.

You do realize there are places where there aren't schools or hospitals?

And as others have pointed out, other departments and agencies (such as the IRS) have most of the rest of the data required to make policy decisions.

There are laws in place forbidding government agencies from merging together datasets.

The last thing people should support is creating of profiles of individuals by combining data from different government agencies. This is why the census is so important as a data collection mechanism.

No distillation. Comparing it to DeepSeek or GLM doesn't make much sense.

Not in the US.

And by your argument poker or backgammon would not be gambling either.

They aren't. A sportsbook explicitly sets lines and you are betting that the line they set is incorrect.

Kalshi and Polymarket are mainly just sportsbooks.

How? They sell contracts between two users. One side each. Completely different from a sportsbook where users are betting that the lines they set are not correct.

A different entity provides liquidity.

There’s no house like in sportsbooks.