Skyy93

If you are implying I am a bot myself i have to disappoint you. I am just not a native speaker, so some phrases could be awkward because I translate german to english.

This article makes no real sense to me.

You think of something new and express it - through a prompt, through code, through a product - it enters the system. Your novel idea becomes training data. The sheer act of thinking outside the box makes the box bigger.

This was the same before, if you had a novel idea and make a product out of it others follow. Especially for LLMs, they are not (till now) learning on the fly. Claude Opus 4.6 knowledge cut off was August 2025, so every idea you type in after this date is in the training data but not available, so you only have to be fast enough. Especially LLMs/AI-Agents like Claude enable this speed you need for bringing out something new.

The next thing is that we also have open source and open weight models that everyone of use with a decent consumer GPU can fine-tune and adapt, so its not only in the hands of a few companies.

We will again build and innovate in private, hide, not share knowledge, mistakes, ideas.

Why should this happen? The moment you make your idea public, anyone can build it. This leads to greater proliferation than before, when the artificial barrier of having to learn to code prevented people from getting what they wanted or what they wanted to create.

A few months ago, I was also skeptical about all of this. However, what we are currently seeing with Claude Code, or Codex, is astonishing: they're building stuff that works and actually improves upon human-coded apps, optimizations, and research. I work in research, and my colleagues are building apps themselves for data acquisition, which usually takes months. They build them in two days. Several companies say their developers aren't coding anymore; they're asking Claude to do it. It feels like an evolution of programming, like the invention of the automatic sewing machine in 19th-century England.

There are many problems and unresolved issues right now. However, we are getting closer. In my opinion, we do not know what we have unleashed.

Do you really think it is still only a bubble? The progress Anthropic did with Claude Code the last few weeks is tremendous.

I first thought this is a porn ad.

Its not really the "ugly" feeling that bugs me. Its the feeling that I actually downgrade when I "upgrade". In my perception an update should make things better, and this is something I do not see with the new update. If anyone has a good example where they actually did something better than before than please enlight me. I would really be interested in hearing something that is better than before.

Perhaps many users think liquid glass as an UI theme does not really improve their UX. At least for me the upgrade was actually a downgrade. The phone got slower (iPhone 13 mini) and I think it did not really add any value. This is of course only a very biased opinion, but I would like to see Apple to react to this tremendous amount of users that simply say no to liquid glass.

Interesting, I did not know that and assumed something far worse from the "delisting". Thanks for the clearification.

Someone posted already the non-paywall version: https://news.ycombinator.com/item?id=46438679

When you bought a game and it becomes unavailable, due to delisting, you care.

It shows a general problem we have with online license's for movies, series and games. You buy the product and at some point later in time you loose access because something changed outside of your control. This is different from VHS/CD/DVD/BluRays where I can use the product even years after purchasing, despite a company loosing the license or simply not existing anymore.

Thank you, thats good to know! I am currently exploring how to process IQ samples. You mentioned its noisy in some frequency ranges, do you happen to know which?

I am currently learning with this book. It has a very good practical and engineering oriented guide, I would recommend it!

Also the mentioned and covered hardware is cheap to get, I am currently using a RTL-SDR from nooelec and for learning the basics you have to invest like 50 euros, which is fair IMO.

We probably won't. GS is a reconstructive method, so when data is unavailable, you can only perform poor interpolation. You would need additional generative, not reconstructive, models. However, this would open the door to unfaithful augmentation again.

Let's just assume the model is a statistical parrot, which it probably is. The probability for the next token is influenced based on the input. So far so good, if I now ask a question, the probability that I generate the corresponding answer increases. But is it the right one? This is exactly where CoT tries to start, in which context is generated you change the probability of the tokens for the answer and we can at least experimentally show that the answers get better. Perhaps it is easier to speak of a kind of refinement, the more context is generated, the more focused the model is on the currently important topic.

No it is not just search. Chain of thought is the generation of new context from the inputs combined with a divide and conquer strategy. The model does not really searches it just breaks the problem in smaller chunks.

Same, ... they did soooo much the last few years with LLAMA or SAM or Lecun's self-supervised methods

In my opinion, you have a very good point here. Why do we want to abolish or automate something that most people enjoy?

I think very few people enjoy laying tiles or connecting a toilet, so a robot in combination with an AI would be the more obvious target for a company. I guess one answer is that it doesn't scale, it's not pure software and the market for tile laying robots is limited. An AI that can easily generate artwork for anyone to offer via a subscription service generates more money.

Here you can find the article about it: https://www.rollingstone.com/culture/culture-features/texas-...

IIRC there was a case where a professor in the US tried to fail students based on such a detector and they were able to prove via Google Docs Revision History that they had actually typed it themselves. It is also stupid to claim to be able to detect an AI that imitates human behavior, of course at the beginning you can still find patterns that repeat themselves, but at some point they no longer do because the imitation is perfect, especially because speech is not a medium as susceptible to noise as sound, for example.

You are aware that these detectors can fail? Its up to you to believe them or not.

This is very US-centric thinking. For me he is just a CEO and completely exchangeable. The only thing I really remember is that the last thing he did was crawl in front of the EU and ask them not to regulate his AI business too much.

He did none of the research that fuels OpenAIs ambitions and future prospects, thats mostly done by people like Sutskever, Radford and many more brilliant scientist.

Than there will be never be a solution to some problems.

Based on Carmacks Tweet: >I am explicitly not working with LLM — there are plenty of smart people doing so. Still, (waving hands vaguely) it does feel like the technology could be used to “argue with itself” and eventually produce a body of knowledge much more coherent than the raw training data.

Source: https://twitter.com/ID_AA_Carmack/status/1630632423351304203...

I would assume they are not

Perhaps there is some belief in this that an AGI can save us from our biggest problems? Energy Crisis, Environmental Pollution, Waging Wars.

But I agree with you ... it is not really urgent since we know answers for the most of the problems but do not like the solution.

Currently the only impact Bitcoin had is burning a large amount of electricity. I would not say this is a particular worthy technology to safe. There would be far better ideas like Transformers in AI or 5nm chip technology.

Yep, but you can do this without acting like an ass.

Naaa, I just wanted to pointing out his attitude towards other research.

As you should know: Schmidhuber even invented the Transformer-Models, to be read in: https://arxiv.org/abs/2102.11174

;)

According to the documentation and discussion about this feature some unit-tests are currently simple skipped: virtual filestem by Emscripten that's not fully POSIX compliant, no processes, no sockets, async only via the browser event loop (source: https://bugs.python.org/issue40280)

This would mean to me they simply do not work or only to an extend. As you can reading the virtual filesystem seems not fully compatible.

“I think there is a world market for maybe five computers.” - Thomas Watson president of IBM

Yes it may be a long way to reach AI that writes good code. But predictions like this underestimate the factor of time.