kromokromo

Yes, that’s the tell tale sign of AI generative pixel art. AI can’t be consistent with the pixels. Looks ok sometimes but never pops.

They are writing these letters to investors. You can read them all as: Our models are so good it’s scary.

Oil will only become more and more volatile. The supply chain is both complex and fragile to geopolitical nonsense. China is betting on stability and reliability, environmental concerns is just a bonus.

High quality photo archive of public bathroom walls with curated text describing the grafitti like fine art would be something I could spend hours consuming.

Kudos to whoever built tiled clouds background on the gcmap site.

Prompt injections hidden into rare books, swallowed by the AI machine.

Model performance is very much subjective to what you’re using it for. I’m a PM and doing mostly knowledge work, and I’ve been really happy with Opus 5 as my daily driver with very good results, perhaps my favorite anthropic model so far. I do a lot of front-end coding with it too. I prefer it over the openAI models, been switching frequently. Kimi K3 is great too.

Im with you, my take-away from the article was of sadness for all the books that are going extinct because of this.

Real humans sharing their unique knowledge, packaged in a book.

No. We need to think in other ways than the ad economy. I’d gladly pay for good content if there was a decent system set up for it.

Its probably just a corporate symptom, weird stuff like this happens in messy large orgs.

Which is often the case in music too actually. So many famous artists and bands created their best work early in their careers. You would think touring and several albums make them perfect the craft of making songs, but I think involving money and external pressure actually kill the creative spark if you’re not mindful about it.

Also a missed opportunity for some of that sweet vanilla html paired with the keygen and crack ansi splash screens

I disagree. They have been winning lately because of better harnesses and interfaces. New actual decent features are shipped almost weekly on Claude code and Claude desktop.

I’ve been juggling between ChatGPT, Claude and Gemini for the last couple of years, but ChatGPT has always been my main driver.

Recently did the full transition to Claude, the model is great, but what I really love is how they seem to have landed on a clear path for their GUI/ecosystem. The cowork feature fits my workflows really well and connecting enterprise apps, skills and plugins works really well.

Haven’t been this excited about AI since GPT 4o launched.

I agree. It feels like people are actually living there, in contrast to sterile, corporate glass/concrete.

Theoretically possible in a distributed way, though usually inefficient. IPFS is a good example.

Asking an LLM to give itself a «confidence score» is like asking a teenager to grade his own exam. I LLMs doesn’t «feel» uncertainty and confidence like we do.

I think a lot of the problems lies in their prompting. AI is usually at its worst when just saying «deploy terraform module». And off it goes spitting out code.

What they should have done as juniors was to have a conversation about the topic and task first. «Help me understand …» learning and planning is especially important with LLM coding.

Bitcoin as a digital store of value is fine. It makes sense to store money in something digital and secure to hedge against local instabilities in currencies, governments etc. People store money in it because it currently secure and it been battle tested, nobody has been able to hack it.

However, I am concerned about the security model on a 5-10+ year perspective. Miners secure the blockchain and get paid for it in bitcoins. We have seen that the mining has become more and more sentralized as the mining gets progressively harder (as by design). Progressively harder mining means that the price of bitcoin must rise in pararell to cover the expenses of the miners securing the network.

Here are my concerns:

- Increasingly centralized mining means a higher risk of a 51% attack (controlling 51% means you control the entire network).

- Bitcoin not able to maintain a constant price growth that will lower the incentive to keep mining as the difficulity increases.

- Quantum Computing actually cracking the encryptions which will kill the entire network instantly.

- Instability in the world might lead to spikes in electricity prices, which will affect the incentive to secure the network with mining.

I also think spending so much energy on security, when there are other viable options available is insane in our current age.

Transitioning to a Proof of Stake model like Ethereum instead of a Proof of Work model would be much more robust with a fraction of the power consumption, but thats just my opinion.

I would suggest containing the chat in a framed container that spans the page, and scroll inside it. In the current form, scrolling the entire webpage makes it loose the chat box effect in my opinion. Right now it just feels like scrolling a html page with horizontal cards.

Opening up coal mines just to bring back jobs in the rust belt does not make any sense. Start mining silicon and other minerals used in solar, batteries and chips instead. It makes a lot more sense even though the initial investment is higher.

100% agree on this. Ive had this thing for 3 years and I still appreciate how good it is. Of course the M4 tingles my desire for new cool toys, but I honestly don´t think I would notice much difference with my current use.

You can tell Teenage Engineering is a fun place to work.

Huge fan of this guy, very talented and creative mechanical lego engineer. You can also tell he spends a lot of time editiing his videos.

It was only a matter of time, with opportunities like that - what did they expect?

Such a missed opportunity though, should have made buttons and sliders physical as well!

To the author: Would be interesting to store all the shapes people are creating and let us quickly browse through them.

Mostly penises most likely (like any "draw something online" service).

(I am also guilty)

From the title I thought you made a optical recognition system for sorting out the laundry, finding sock pairs etc.

Very cool sim, but kind of disappointed still. Will build this someday.

This is just doomerism. Even though this model is slightly better than the previous, using an LLM for high risk tasks like healthcare and picking targets in military operations still feels very far away. I work in healthcare tech in a European country and yes we use AI for image recognition on x-rays, retinas etc but these are fundamentally completely different models than a LLM.

Using LLMs for picking military targets is just absurd. In the future, someone might use some other variation of AI for this but LLMs are not very effective on this.