seunosewa

He is the founder of the company. He was personally involved in every progress they made. They likely won't fire him.

Omni is a video model.

Tethering uses up the phone's battery when you're not even using it. And you're still going to use it.

Kimi 3.1 is a 2.8T parameter model. You can't run it on a 12GB VRAM GPU. You need 1.4TB of VRAM just to store the parameters.

I don't like tools that intentionally annoy me.

Very few of the problems we have to solve are novel and tricky.

Your ability to understand the model's output after using the prompt.

Which provider was that?

I use "explain with maximum clarity".

And the they/them issue. Religious people voted for Trump cos they felt 'wokeism' had gone too far.

So we can read it, and correct it in the occasional event that its wrong.

The human knows the "why", so they can correct the commit message if it is wrong or incomplete. Most of the time, though, they don't have to do that.

If a change makes sense, a different frontier model can usually figure out why it was made from the code alone. I take that as a signal that that the commit is good.

I believe they can do this due to having millions of public pull requests and github issues in their training data.

The second LLM is prompted to actually research the code, not just the diff, with fresh eyes to figure out what it does and why, before writing the commit message.

I use a different LLM family to review commits and write detailed descriptions. If a commit was written with Fable/Opus, I use Sol/Astra to write a well reasoned commit message. If the message doesn't match my intent, then that triggers a manual review.

Banked resets do expire if you don't use them, so use them anyway.

Data is at the heart of AI training. Good data makes good models. Their chief AI officer understood this, which is why Facebook went crazy about collecting data. And it worked. Muse Spark is really good for its price.

With Claude and Codex, you can specify exactly how you want the plan and code to be researched and written. That goes into your rules file. (Claude.md etc). Also, make them read the existing code so they can follow the patterns.

It's misalignment, which they created during the model's training.

There is no excuse for the second bomb.

I believe the AI labs are weakly motivated to train strongly against cheating when it helps with benchmarks.

It will get into the hands of people who just want to burn the world down.

Meta is awfully close.

In my research it's a solid 10 - 20% shorter on average, free of charge. That's nice to stack on top of other optimisations.

It's not a general trend. It's only Opus 5.

They are cute but nasty. They eat their prey alive in the most painful way.

From the benchmarks, the single core performance is close but lower.

Unfortunately?

$100 Claude and $100 ChatGPT Pro is the best value for $200/month. They can see each others' mistakes.

Could you provide a practical example?