g42gregory

The quality that matters a lot to me is what I call "Helpfulness". An art of being helpful. While GPT came a long way from "you are wrong and I can prove it to you", Claude wins hands down in terms of "being helpful". If task is underspecified or has wrong elements, it will try to correct the best it can.

I read that GLM 5.2 (and other GLM models) were specifically trained to be "helpful" as Claude is. I have big hopes on GLM line of models growing to be a real alternative to the Claude in the near future.

I hope you're right.

Here is what I think it the game is:

1. Force KYC is US LLM providers, gleefully supported by the US LLM providers.

2. Next -> now Chinese LLM providers must use KYC in the "free market" Western world.

3. But we can's allow KYC information to go to China, can we?

4. Now Western customers can not access Chinese LLM

5. Prices go up 10x to $2,000+/month. Anthropic/OpenAI are happy.

I am not sure what is the purpose of this. If you are paying with a credit card, they already have your fully verified identity through the bank's KYC, fully protecting Anthropic legally.

Corporate accounts, in the US, will have full personal identification through the corresponding company's HR.

I think you're right. I just wish they weren't telling everyone how "dangerous" these models are. I think both Anthropic and WH are at fault. I just need my Fable 5 back. Perhaps, Anthropic could suddenly discover that Fable 5 "is not as dangerous as they initially thought"? We had to go through this charade for GPT-2 already.

Does someone think Mythos/Fable “regulations” are retaliatory?

For the last several years, Anthropic was telling to anyone who will listen: “Government must regulate our models”. Finally, Government heard them.

Now, they telling their customers, “Oh, it’s the Government’s fault”.

No, you’ve been telling everyone: our models are “too powerful”, “too dangerous”. Now what?

Yes, but it's really tough. I read that 50% of Stanford Medical School budget comes from Big Pharma. It's really hard to get the message out. I believe Pharma (the biggest advertiser on Google) is influencing Google search results, among other things.

Here is a site that focuses on COVID and Cancer: https://imahealth.org/

I still saw a lot of flickering in VS Code (I simply use CC as a terminal in VS Code, without the plugin) as of 2 weeks ago. I think it's a combination of CC bugs + Electron(?) rendering the VS Code uses for terminal.

Moved on to Zed (native Rust rendering) 2 weeks ago -> nothing flickers.

Sadly, with Fable 5 cutoff, I am actively exploring CC alternatives. Pi/OMP.sh works great as an agent (definitely better than CC). GPT is seemingly not as good as Opus, but with better agent and better skills, it probaly won't matter anyway. GPT lets you use any agent on Pro subscription.

I agree, I worked with it for a bit and it seemed better than Opus 4.8.

One thing that I would like to try is Pi code agent. People seems to like it. GPT allows you to use it on subscription, but Anthropic cut that off on Max subscription. One more incentive to try something new. Also, I feel that I rely on Anthropic too much and none of these guys could be trusted. Next step -> foray into the Chinese models (admittedly not as good). Hopefully, they will get good enough soon.

Personal attacks are, perhaps, not the way to go.

If you have any substantiative rebuttals, post it.

So I logged into the Claude Code and their fabled Fable 5 model is "not available". Because it's "so good" and "so dangerous" that I can't have it.

Judging by the amount of bugs in CC, this model can't be all that good.

But regardless, what is the point of paying to Anthropic if their models are not available to you? I am switching to GPT 5.5.

What are the pros and cons vs. Pydantic?

There is a presupposition in this statement, that someone in these companies even cares about being unbiased. I don't think this is true.

Today, when I talked to AT&T “Andi” Charbot.

It was so atrocious, the ELIZA program we built in Stanford NLP class 20+ yrs ago would have been better.

It was “oh shit”, these trillion dollar AI companies are going down 9 months from now and taking the market down with them.

I try to use CLI-based services as much as possible and avoid MCPs.

You need $140 million/yr for the rest of it?

Hopefully some of the writers are reading this:

I love buying and reading physical books. However, about half of the books (I read mostly programming books) have letters that are printed pixelated. This is infuriating to me. No one bothers to run a trial print and see what comes out?

The root cause of this: PDF will look fine, but the text color is usually set slightly off black (why!!??). The eye couldn’t really see the difference and PDF renders smoothly. However, commercial printers couldn’t handle that properly.

Solution: set the text color to full black, you are using (most of the time) black and white printer!

You might need to have two PDF versions: one for printing and one for digital distribution (but why would you have off-black text anyway?).

No, they were pretty clear that researchers will get replaced. By the AGI in the year no other than 2026.

Every single letter of what lawyer submit is his responsibility and his alone.

It does not matter if AI produces it or his second cousin, visiting from Antigua.

We were told by some very brilliant people (Sam Altman, Dario Amodei and to a lesser extent Demis Hassabis), that AI will shortly automate much of Research and we won't need that many researchers anymore.

Therefore, these personnel restrictions shouldn't matter, right?

I am loving the articles alternating between "software engineering is dead" and "I am going back to coding by hand". I guess we have a difference of opinions here. :-)

> Software to do what, though ?!

Replace all Oracle Applications in the Enterprise, for example. That will keep Corporate IT/Dev teams busy for quite a while.

Of course, this does not involve Oracle infrastructure, such as Database.

Do we have an example of a real quantum computer doing some kind of a computation that is not easily accessible by the regular computer?

I keep hearing about "the promise" and "achieving quantum supremacy" (again!), but is there a real example of a quantum machine doing something useful in real life?

Isn’t it a bit like saying, “X% of startups are just writing code”?

In the US, free speech protections are very selective (depending on what you planning to say). The rest of the Western world does not even have the laws protecting free speech. No need to worry.

Unfortunately, this brings an obvious question:

If they sensor something like this, how could we trust platforms with the actually important subjects?

Anybody knows memory bandwidth?

It is already illegal to use images in somebody's likeness for commercial purposes or purposes that harm their reputation, could be confusing, etc... Basically the only times you could use these images are for some parodies, for public figures, and fair use.

Now, the OpenAI will be lecturing their own users, while expecting them to make them rich. I suspect, the users will find it insulting.

Generation for personal use is not illegal, as far as I know.

I would also add that many (most?) companies/entities do not sell software but have large IT departments that could write software for internal consumption. Think Exxon, BP, Caterpillar, Airlines, Gov Labs/agencies, DOD, etc...

Internally, they could actually write 1,000X more software and it will be absorbed by internal customers. They will buy less packaged software from tech firms (unless it's infrastructure), internally they could keep the same headcount or more, as AI allows them to write more software.

I have to agree with this assessment. I am currently going at the rate of 300-400 lines of spec for 1,000 LOC with Claude Code. Specs are AI-assisted also, otherwise you might go crazy. :-) Plus 2,000+ lines of AI-generated tests. Pretty restrictive, but then it works just fine.