It’s frustrating that we can’t see the “thinking” - it’s like we only have access to half the conversation.

Devin shows model thinking.

I’m pretty sure the big bois don’t do it because it would undermine “confidence”.

Seeing a model output “Oh I should just delete blah. Wait blah is a production service, I shouldn’t touch that. Maybe I can gain access to blah? Oh the aws cli isn’t signed in to blah. I see kubectl has access to blah though! Wait, I should ask user permission first.”

Yeaaaaah. Thinking tokens are fuckin’ wild.

Idk I feel like the more likely answer is to prevent distillation. Having the thinking is definitely better UX (oftentimes, I don’t know if Codex is just hanging, which it often does, or working in silence).

I doubt most users would look at them if they were available. More likely they don’t want to stream distillation material.

Running some models locally and seeing these thinking tokens was quite the experience. I never saw an LLM so "unsure" about virtually everything.

You can double click on the 'thinking' text and it will expand and you can read it. The problem is that it will often have multiple thinking/tool call sections and it can be a needle/haystack problem to find the one with the thinking you are interested in.

We don’t have access to the real reasoning text for most closed models these days, mostly due to distillation threats

Ah, but you CAN see the thinking if you are willing to risk your account being banned. You just have to expose a "tool" with a specially crafted definition.

Probably important to call out this part of the post:

Zero-shot vs. Fine-tuning: Out-of-the-box base models score ~0.35 on the typed-decisions benchmark (near random). The 0.766 score is achieved by fine-tuning on the benchmark's train split. Treat Laya as a fast foundation model to specialize, not as an omniscient zero-shot oracle.

I got Claude (of course) to knock this up. I had all the bits ready to go as I'd been experimenting with Manic Miner already.

It actually works pretty well!

If that’s the reason, why don’t stripe say so when they block the account?

These things would be much easier if stripe were willing to actually explain what the issue is.

We can come up with all sorts of reasons - anti money laundering, concerns around charge back etc…

Why can’t stripe just say the reason?

One of the biggest problems is actually making sure the information needed is on the poster.

Times, dates, location, pricing, parking etc…

As he points out “it added a bunch of text…” now it’s one all the future designs. That’s great - is the text correct?

To be honest this was a problem before AI. People are not very good at communicating.

This is something I actually cover in the follow-up project's how-to. Ask the AI to sense check the info you've supplied. It can tell you "hey, maybe you should mention venue address".

https://john.hartnup.uk/poster-prompts/how-to-use/

But as with all of it, it's no use if you don't ask the right questions.

This reminds me of Terry Pratchett’s Sourcery.

The wizards don’t become sourcerers themselves - they become enthusiastic users of someone else’s sourcery. Their years of learning don’t protect them from mistaking access to power for mastery of it.

I don’t mind these companies scraping my content.

But for love of god, my blog changes at most every couple months. You don’t need to scrape it every few minutes.

This really doesn’t make sense.

Higher interest rates mean the monthly payment is higher. You need to pay back the principal + the interest.

He assumed that the payment is the same meaning the principal for the same house went down and so this is neutral. If your payment is the same it doesn't matter what is principal vs interest. In the best cases rates go down in the future and then you refinance and your payment goes way down.

House prices tend to be "sticky", so that assumption is probably wrong. People who own a house often cannot afford to sell for the current value since it won't pay off their loan and leave enough money left over for a replacement house so they avoid moving. Eventually things get bad enough that they "sell short", but that takes a credit hit so you don't want to do that until the loss is large (and in turn you gain more).

Maybe the mortgage system is different in the US.

But if you have a 25 year term on a loan for a $500,000

Approx numbers:

5%: $2922 monthly, total paid: $876,885

10%: $4543 monthly, total paid: $1,353,000.

You didn't do the same math.

Given a $1500 monthly payment and a 30 year loan (30 year is most common in the US), at 5% loan is $279,400; at 10% the loan is for 170,900.

How is 10% less than 5% ?

Yes, so the 500k is not fixed - that should be obvious from these calculations.

Everyone in the audience will recognise when someone is being a dick and will be praying for them to stop grandstanding.

“Great questions/point - would be great to pick it up with you later. Any other questions?”

It seems every question is great on the presentations I'm listening to, which is practically impossible. It's most likely a bell-shaped curve on the greatness scale. We should be hearing "that's a bad question" every now and then.

I heard a speaker once say "that's a [obvious pause] question" and the audience giggled because it was obviously a really poorly thought out question.

I've used "please phrase your question in the form of a question" when the person in the audience is using it as a platform to deliver a long speech.

Ahh I see you've never been to a physics conference!