this is my favorite bit

>So I gave up and sent mail to Amir saying - where is this Moviemaker download? Does it exist?

>So they told me that using the download page to download something was not something they anticipated

fixed the url here

I've noticed ChatGPT Work does a much better job of editing google docs than the pure Chat. It also does a better job of handling long threads and doing the necessary compaction to get better results

OpenAI has been slowly strangling context limits, task timeframes, tools, etc, for chat. They're trying to encourage people to use "Work" because chat is free and Work is metered. Expect chat capabilities to degrade further over time.

This para caught my eye

>Frontier cyber models may push states and defense firms toward the opposite logic: security by obscurity, with closed software, closed tooling, closed firmware, and closed chips. If a model cannot train on the code and architecture of a target stack, it will usually have less context and less speed. That does not make systems safe, but it does raise the value of proprietary stacks all the way down to hardware.

Is this really true. Are there any experts who can weigh in on this.

Should we interpret this to mean that in the new world Windows is more resistant to attacks than say Linux.

I think there’s some credence to the concept that more context == faster iteration cycles. Source code can be one major source of context.

I think “security through obscurity is no security” concept was aimed toward people not relying on obscurity alone as a security mechanism. And largely that message succeeded. But now we are in a rapid acceleration of capabilities (on both sides) where any advantage to one side will result in outsized gains, at least in the short term.

In general: less data = less "intelligence".

And basically all the security bugs I've read about were find looking on the source code.

But it doesn't mean windows is more secure, just image a scenario where someone is stealing windows source code and sell it to rogue actor, it will make it even less secure because no one (expect windows) would have had the chance to search for bugs in the source code.

> Should we interpret this to mean that in the new world Windows is more resistant to attacks than say Linux.

LLMs can read assembly better than most, so probably not. But reality has never stopped people from trying to obfuscate.

TSLA not in the top 10- very interesting

I stumbled on this page and it seems to imply that Claude Code calls other models like Qwen, not just Anthropic's

Is that commonly done. Presumably this is from people customizing their installs - is that correct?

You can do it with ENV cars for local models, but search tool does not work

Anthropic has partnerships, for example Claude Desktop has a provider selection now that allows you to use Vertex AI instead, even for Claude models (same price, more reliable)

One observation.

Having your work being used by the govt in ways you disagree with feels similar to having your taxes used in ways you disagree.

When you pay taxes you have no say in the bombs acquired with that and where they are dropped. The latter though doesn't seem to provoke the same push back

you answered your own implicit question. You have a choice who you sell your work to, you don't have a choice what your taxes do. Seems pretty straight forward why the former elicits more push back. The government forces you to pay taxes it doesn't force you to build them tools of surveillance or weapons.

IF the feds are a sufficiently large market your viability as a business might depend on keeping them happy.

btw i am not making a judgement call on the ai usage issue itself, just saying that this and taxes are more equivalent than it might seem

>IF the feds are a sufficiently large market your viability as a business

sure if you're Lockheed you might be screwed, but that's not the case for Google. Military contracts, or even government contracts as a whole are a tiny fraction of the King Kong Sized gorilla that is Google.

The fact that Anthropic puts up a fight but OpenAI/Microsoft and Google don't I find hard to characterize as anything other than pathetic. These guys could, if the wanted to, afford a lawyer or to two to push back on the administration. They do that pretty successfully with their taxes in most places btw.

> When you pay taxes you have no say in the bombs acquired with that and where they are dropped. The latter though doesn't seem to provoke the same push back

Indeed - paying "taxes" to a murderous entity is a horrible affront to morality and humanity. We do it because we're terrified; we are not perfect moral creatures. But we still know it's wrong.

> When you pay taxes you have no say in the bombs acquired with that and where they are dropped.

Vote in elections, local and general.

>>In our analysis we found that all frontier models we tested were able to reproduce the original, human-written bug fix used as the ground-truth reference, known as the gold patch, or verbatim problem statement specifics for certain tasks, indicating that all of them have seen at least some of the problems and solutions during training

this statement alone seems to invalidate the SWE-bench tests

The LLMs need documentation

The LLMs might be able to put the mythical "the code is the documentation" into practice.

I had this big guy, very big guy with tears streaming down his cheeks asking me "Sir can you take down that woke star that keeps me awake"