There is no FUD here, just irrational panic in our markets.

There was no lie. It was clearly understood that they cited power costs. They never said they spent $5.6 million to buy hardware. Why blame them for the lack of technical literacy of readers?

"No wireless. Less space than a Nomad. Lame."

Oh Slashdot! Never grow up!

ChatGPT, the fastest growing consumer application in history, isn't doing it for you? Not prime time enough?

> In the coding-space anybody that does work on that level is probably just doing interfaces, no real code.

Very few programmers do "real code". Most code outside core components is just a tedious artifact, interfaces or not.

> The most important thing that could be added to LLMs is a fact-checking ability. But that is not actually possible mathematically. It would have to be a completely different tech and none of the known approaches can do it outside of _very_ narrow subjects and pretty simple facts.

You just ground it with RAG.

I used abstract language. I understand that encourages subjective interpretation.
Those are not the issues I have.

These are research and niche projects, not huge amounts of code, no legacy code, no team, no evolving business customer requirements to deal with and I propose most of the feature set - so you might say these are simpler projects than you might have. But these are significant projects and the stake holders are important people. I have been programming for decades, although my job description was never that of a programmer. I find LLMs to be fantastic for my workflow. I would not go back to the world 2 years ago and I constantly get better at applying them to generate code, just as they keep getting better. They changed my entire approach to programming.

It already makes me several times more productive. The next step of evolution is in inference-time scaling with agents. That will make most of the current app development workflow disappear. You will still need programmers to be responsible for managing the business specific logic, but the framework code should melt into background. That need not mean we will need any less programmers, it just will mean programmers will largely focus on a different set of considerations and have a very different set of expectations in terms of productivity. In the past, I would not do many advanced things in code because I did not want to introduce complexity. Much of that is trivial to iterate with AI now.

Most people who hire programmers today don't look for your skill in how well you manage your memory and how optimized your code is because we have long moved into high level languages with garbage collection. Skill in writing code will be the equivalent of skill in writing machine code today. It's valuable for a small set of developers, but most developers are not hired for that skill any more. Most software projects are not original. They are just tedious.

Or looking up what an oligarchy is
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fgizmodo.com%2Fa-lot-of-a...

It's not the best. Qwen is better and so is Deepseek. They are all open weights.

None of these are first, second or even the third best. So the only way they can remain relevant is by being open weight models. If Llama ever out does everyone else, I doubt it will remain open.

All the other features you mention around GPT-4 (web search, code interpreter) have nothing to do with the model, they are quite simple features and all other models can do them as well. Several open source implementations exist for these features that allow you to plug in any model.

That argument is like saying sometimes cars or planes spectacularly crash (and here are some examples), therefore you are a fool if you bet on them.
This kind of anecdotal reasoning makes little sense: Technology isn't going to stand still and the goal is not perfection, just being better than above average humans (or lower tier experts) on average is.

Do you want me to show you how misapplied medical treatment caused death? I cac show you way more than 12 examples. Will you stop using all medical care?

> And the more we use it, the fewer empty chambers there are in the revolver.

How so? We get better at it as we apply it.

> no matter how much you get paid to shill for it

Wow, just wow. You just assume I am might be getting paid to post this? Do you realize how unhinged this is?

> Failures are easy to find, and well known.

AI does not need to be perfect. It just needs to make errors less frequently than a human. If I can have an AI that only makes 10% fewer errors than a human, but is 1000x faster and 10000x cheaper, would that not change our business processes?

Both AI and humans will always make some errors since each of those tasks is not perfectly specified, that includes medical diagnosis. It's just a matter of how often.

And if it isn't "good enough" today, given the rate of progress within the last couple of years, are you expecting that to hold for long?

You sure about that quote? No hit for the quote on Google for that.
Wikipedia: There are 3,000 castes and 25,000 sub-castes in India

US healthcare has 80 hour work week residents. Yes, disasters happen, but its here was well.

Not even close.

AI as unlimited use cases today. Every single one of us will be using AI backed tools and services, whether we choose to or not.

Neither is a "scam". One has matured to usefulness, other also will, just has not yet.

I haven't used Qwen-Plus, but it's amazing what even their instruct and code models can do at just 32 billion parameters.

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fqwenlm.github.io%2Fblog%2F...

2025 will be exciting for those with interest in running these models locally.

We no longer look to Open AI, which is pretty much the most Closed AI at this point. It's quite surprising to see China take charge in the open weight model space this quickly.

Qwen models surpass Meta's Llama models. Microsoft's Phi-4 is supposed to edge it, but has a track record of its real world performance not matching benchmarks in the previous iterations.

DeepSeek is open and cheap too, but they have not added much on the consumer desktop scale since their last lite model.