←
Mike Veerman9,372 posts
Mike Veerman avatar

Mike Veerman

@mikeveerman

Hands-on technical leadership in AI. Building FALKON (getfalkon.ai) We Should Do Something With AI newsletter: somethingwithai.substack.com

◷ Joined April 2010▧ 283 Photos and videos
Mike Veerman avatar
Mike Veerman@mikeveerman·11h

There will probably be big business in selling "better niche intelligence" Sure, Claude can draft a meal plan. But there might be demand for an endpoint that generates dedicated, high-quality meal plans for diabetics. ChatGPT can generate a contract. But a bespoke API will beat it by generating a real freelance agreement for the German market. None of these are full-fledged products. They are also not trivial to build. They are integrations that products or agents can use. I bet people are willing to pay for that.

Zack Korman avatar
⟳ Mike Veerman retweeted
Zack Korman@ZackKorman·16h

This is the correct take. I absolutely can't believe we are building AI that goes "sorry I can't let you do that" and then acting like that is somehow a safety feature

AMD avatar
⟳ Mike Veerman retweeted
AMD@AMD·Nov 17

We’re excited to share the completion of our @taalas_inc acquisition and welcome their world-class engineering team to AMD! Their technology strengthens our full-stack AI portfolio and long-term inference roadmap with breakthrough performance and efficiency for increasingly specialized AI workloads.

0:09
Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 17

Pi: "Honest take: finishing the entire set of cases will be a very long grind, potentially taking days. Are you sure you..." Yeah, buddy, ping me when you're done next week.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 13

Abundant intelligence needs to be open source. It needs to be "for all". It can't be the monopoly of a few mega-corporations.

DHH avatar
⟳ Mike Veerman retweeted
DHH@dhh·Nov 13

Local AI. Local compute. Open weights. Open source. Global collaboration.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 11

"This used to be amazing, but now it's just OK" Maybe what we call nerfing is just us getting used to the model's capacity.

DHH@dhhNov 11

I'm starting to think "nerfing" is the evil spirits of the AI world. Every rustling of the leaves, demonic. Every prompt response not to your liking, nerfing.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 10

I'm all for using AI to cut bureaucracy, but this is not how recruitment works. This guy makes hilarious content, but the next one will just run his own bot to get to the next round.

I cannot even 😂😂😂 Human (guy with shoe on head) trolls AI Karen Bot in an interview for a minimum wage job at Burger King. Absolute Perfection. 🤣🤣 Humans for the WIN!

3:45
Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 9

"Why would they give it away for free?" Open source is a distribution model. It is THE distribution model.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 5

The difference in sentiment on X vs LinkedIn is staggering. X is full of engineers that embrace our changing industry. On the other platform you get swamped with negative replies by devs who have Open To Work in their banner.

There is a subset of engineers who have decided to wait out the coding agents wave and still expect the industry to return to typing syntax.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 4

ChatGPT was launched at the end of 2022. We've been coding with LLMs for at least four years now. Yes, it's moving fast, but it's no longer brand new.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 4

There is a subset of engineers who have decided to wait out the coding agents wave and still expect the industry to return to typing syntax.

Mike Veerman avatar
Mike Veerman@mikeveerman·Nov 3

While some people are still discussing "reading the code", LLMs are manipulating weights in neural networks to increase their math proficiency. Do we need to find ways to test and verify that? Of course! Will that come in the form of reviewing the changes to the matrix? Good luck with that...

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…

Post attachment