beering

I commonly see people online say something along the lines of, “instead of curing cancer we got a stupid chatbot.” Dario is simply responding to what people say they want.

It’s an econ paper. What do you want? A page of Claude slop with the small caps subtitles?

Gotta build some personal benchmarks if you don’t trust the public ones. But the well-known public ones, despite their flaws, are generally high signal on model intelligence.

You (or anyone else) can just benchmark and compare. If they were serving a dumber model it would be trivially detectable.

You are right, reasoning is unrelated to tokens per second.

I’m so excited for driverless cars, and I wonder how this will affect the shift towards services like Waymo. If the choice of whether to take Waymo vs Uber comes down to cost, unions seem like they would shift the usage towards Waymo. (or Waymo simply increases prices to match)

isn’t strong sandbox a good thing?

Consumers in Germany generally understand that this is a cost-savings device, not a backup supply of electricity. You tell your neighbors and friends to get it because of the energy bill savings.

They are not as weak as the toy panels you are describing, but they are regulated to 800W max anyways.

AI slop webpage. Also none of these “anti-AI” clothing projects tell you that they only work for one or a few related CV models. They are good clickbait but not real products.

You could start a new business selling grindstones to the axe grinders. Like selling pickaxes in a gold rush.

Number 3 sounds like nothing out of the ordinary? Any bonus you get from an employer, granted or otherwise, has to be taxed at some point. This just sounds like a weird signing bonus.

It’s interesting how people will interpret news in ways that seem bizarre, but they were looking for something to be angry about.

If you remember how crazy verbose previous gpt models were… clearly there’s something going on unrelated to cost. It would restate the same thing in different words several times and fill the output with emoji or lists.

Luna correlates with Claude Haiku, not sonnet. I think Terra would be the equivalent to Sonnet.

Yes, it was a joke he made about how loyal his voters are, and it was widely covered in the news.

Training the LLM to do things that the user didn’t explicitly ask for is a good way to get complaints from the users. Doesn’t matter if those things are best practices.

If you were a mathematician and came up with any of these results, people would pay attention even if you published it on your blog. What’s the requirement for needing to publish it in a journal? OpenAI is not trying to achieve tenure.

Almost everything I’ve learned in school is learnings handed down from others, not things I discovered. Is all that knowledge useless?

And no, science is not a branch of philosophy and not everything is a philosophical question, despite what the philosophers like to say.

If everyone publicly said that the models can only do things that humans have already done, but you know they can do more, wouldn’t you want to show them otherwise?

Math ability also helps with other things like making models more efficient.

Solving math problems doesn’t require the cooperation of rival factions.

Follow up: I can’t seem to find any ruling that says you need to destroy the scanned book. Destroying the book might bolster your case in a copyright lawsuit, but the law does not require it afaict. I doubt you would lose a lawsuit simply because you put the book in a locked warehouse after scanning it.

Is that true? I don’t remember Google Books destroying scanned books and they scanned a ton of books.

Partially agree: yes we should endeavor to learn as much as possible about how these reasoning strategies work. It will pay dividends in enhancing and aligning the models.

But the stochastic game IS the win. That is exactly why they are able to find solutions is seemingly infinite solution spaces. Your symbolic techniques can only get you gains in narrow domains and by the time you figure out how to make it work for your niche domain, the next all-purpose LLM release will crush your results with stochastic games. (OK maybe over-exaggerating a bit here but these stochastic games over the language space is why we can pull together knowledge from many domains.)

Some people might be on the ChatGPT Pro subscription plan or consuming their employer’s tokens.

The company losing money does not mean the model inference in API is 70% subsidized- that’s a crazy leap in logic. Obviously the massive number of free chatgpt users are getting subsidized 100%.

You really really don’t need to pick. Just use Sol on high. That’s my daily driver and I don’t touch the model picker at all.

Now, if cost is your concern, then that’s a problem in all of computing. Hence why I’m sending you short plain text messages using an iPhone with a many-core CPU and gigabytes of RAM.

True, and the pricing plus token efficiency seem to suggest that gpt-5.6 is much more efficient. But hard to know since neither lab publishes that info and it’s a bit of an apples and oranges comparison.

Prop up the existing wage-slavery system because anything else would cause people to not have healthcare coverage? It’s shockingly unimaginative.

Anyways, my last Uber driver was playing games on his iPad while driving. Did a good job of hiding it at first.

I looked through the report. It’s a bad report but it couldn’t have been done well because of not having apples to apples data. They compare data from very different environments (all CA Waymo vs NYC taxi) and just kind of shrug aside that difference.

Really awful how the AI labs are skillmaxxing /s

Pelicans aside, we need to remember that benchmarks are the only good quantitative way we have of comparing models. If someone has complaints about “benchmaxxing”, please ask them to contribute a better benchmark! It is valuable work and very appreciated.