vorticalbox

Has anyone tried training a diffusion model for speed ?

not sure if you will see this but i see what you mean. i have been using it the last few days there is just something about thats just "nice". its not the fastest on my m4.

Just tried this on sonnet, it believes the developer documentation actually publish these as a way for developers to test refusals.

After checking the document it doesn’t publish or even mention these.

Is this was xai isn’t winning because they still use numbers for grok releases?

Yes this has been very useful for me this week. I trapped my middle finger on my right hand in a door and it really messed up my typing.

I tried using dictation but I really disliked speaking to it.

You can write a really messed up sentence and it almost always worked out what I meant.

Quite amazing actually

If I remember correctly I read a paper a few years ago that tested this with people learning piano.

The control just did their lesson and the second after the lesson sat and just imagined them playing what they learned no finger movement allowed just thinking.

The second group over all learned faster.

Did you happen to try bionic?

https://lmstudio.ai/

Could this not have just been a skill?

I downloaded Muse today actually to try too. Seem it said a few times on here that’s it’s good.

been playing with LFM2.5-24B-A2B on my m4 mac, 700-900 prefill and 60 t/s in generation, its doing pretty well writing some python scripts for me.

I remember a long while ago when you could jailbreak iOS with a corrupted font in a pdf.

There was a website jailbreakme you simple slid a slider and you were done.

From Wikipedia

August 1, 2010, exploited a vulnerability in the FreeType library used while rendering PDF files.

https://en.wikipedia.org/wiki/JailbreakMe

I assume your still verifying that opus is correct?

In which case congratulations you’ve just been promoted to manager with no pay increase!

Joking aside Ai is doing a large chunk of my job but all it’s done is shifted what I actually spend time on.

It’s not really saved any man hours.

Are there plans to release the detector?

I didn’t wonder why the bench marks were done in medium. The medium pelican is better than the high.

Maybe reasoning negatively effects the output after medium.

Except it’s not accidental? OpenAI set a program to do a task knowing full well that it can and does (and often it seems) hack into external systems.

I don’t see this as any different than say a user of sqlmap accidentally putting in the wrong target.

I have also used ramda for this

R.pipe(f1,f2,f3)()

I got to 2/30 then it just guessed 7 from 30-50.

Did I break it?

Have you considered contributing upstream?

Isn’t this what docker is for?

Jokes aside I do the same I tend not to upgrade my Mac, my phone unless I really have to.

The “update” button on the top of my brave install has now turned from green to orange and I’ll likely wait until it goes red

I saw a post on here, basically they get ai to design the change and give them each each one at a time that they manually type in.

They argued this still allowed them to fully understand what has being implemented and how it fitted together.

I’ve been doing this since I read that and it also allows you to catch stupid stuff while your typing, you can reason about what each little change does and why it’s needed.

This also lets the LLM change the future of the plan if you fine something.

This turns out to save a whole bunch of time later because you already know how it works.

It’s not nearly as fast as just letting the agent do everything.

I had an agent working on a bug decided the fastest way to fix this was to write one off javascript scripts that loaded the database uri and query the data.

I mean it’s not wrong but I was like maybe not do that without asking first.

It’s immutable npm, you can’t unpublish I believe but yeah it’s trick of allowing one to publish typescript packages rather than compiled js only applies to deno.

Actually not tested that I will give it a go. I usually keep reasoning on low.

jsr is moving to cloudflare.

My work pays for Claude and cursor.

I have actually just dropped to using sonnet for everything, sure it does need some directing but I have yet to see a need to jump to opus.

To me it feels like sonnet/terra and composer 2.5 and grok 4.7 are actually good enough for most tasks and these companies are pushing the high models simply to make money.

Reading your comment has just brought to mind the sheer amount of betting and stock trading ads I get on YouTube currently. I’m in my late 30s so a little out of the “young males” category.

The issue is “ai generated code” using ai to find bugs/0 day and manually writing a fix would (I assume) be allowed under the new rules.

You can sort of do this.

For a given bug one could write a test that prove its existence, this gives the LLM a target that they can actually iterate towards.

When I was a teen some 20 odd years ago the “algorithm” is people you spoke to.

That said smart shuffle on Spotify has found me a lot of good music.

I installed this when I first saw it on HN.

It’s great that’s plugins all the way down the issue I find is I haven’t missed anything enough to need a plugin.