Hacker Newsnew | past | comments | ask | show | jobs | submit | zius's commentslogin

In my first interaction ("hi there kimi k3!"), Kimi K3 identified twice out of three times as Claude:

> Hi there! Quick note — I'm actually Claude, made by Anthropic, not Kimi. But no worries!

https://imgur.com/a/jqpc2Jc

and

> Just a quick heads-up — I'm Claude, made by Anthropic, not Kimi! Kimi is a different AI assistant (made by Moonshot AI), so it looks like there might be a little mix-up.

https://imgur.com/a/AKxeysH


FWIW Claude sometimes identifies as DeepSeek when asked in Chinese: https://x.com/stevibe/status/2026227392076018101


I remember when Claude identified as ChatGPT a long time ago. It proves nothing else than that there is a lot of training material on the internet with Claude as the AI.


I don't understand why labs don't include correct model name in the training process. Almost nobody seems to be willing to tell their models who they actually are.


If the model works correctly, you don't need to include the model name in the training process. You just add in the system prompt "You are X, trained by Y" and the model will claim to be that. That also allows users to "white-label" the outputs sort to say.

This approach is basically how all "models know who they are" (they don't actually typically "know" that at all), it's just a system prompt instruction in the platform you use.


But I would expect they would at least use post-training to discourage identifying as the wrong company.


A lot of API customers don't want their chatbot or agent to identify as Claude or GPT, they want it to identify as <customisable marketingname>.


That's probably the true reason.


That only makes sense if you don't sell access to the model for 3rd party product development. One of those products is chat personas/agents. The company integrating the model will want it's own identity.


You can put anything in the prompt. And if you have bare model that you don't give any prompt from the start and want to find out which model you are talking to, you are out of luck.

I think Qwen teaches its models that they are Qwen. Most others don't bother.


Because they do not know the name of the model before they train it. There is also distillation, where multiple models will be trained from a larger one. E.G. Sonnet was promoted to Opus at one point after it surpassed expectations.


How many times do people need to point out that every model has this behavior until this stops being posted?


Models don't have an inherent identity. It should be obvious by now that every models trains on public AI chat session transcripts. I've seen Claude say it's Qwen.


This happens with other models too - Gemini often identifies as ChatGPT for me, confusing many a debugging attempt


It's distillations all the way down!


Models don't know who they are. Stop asking this question to the model as a source of truth.


The meme/trope of China copying everything really keeps playing into itself


Could be on purpose to disguise as a US made model


Bootleg AI



I can see how, if all you ever were exposed to was that link and this post, you might think they are equivalent, and the conclusion is "whatever, everyone is stealing from each other.

But the answer is no, Claude didn't train itself on ChatGPT, but Kimi K3 did train itself on Claude.

There has been no accusation that I am aware of by OpenAI against Anthropic, (which are in the same jurisdiction, so it would amount to legal action). On the other hand there have been accusations with extensive detail by Anthropic on how chinese models are attacking Anthropic to reverse engineer and copy their product.

There's nuance if you care to see it, but maybe it's easier to pretend that everyone is ripping each other off so that you can consume a ripoff without seeing yourself as at fault.


point is everyone scrapes data from the internet. Do you think there is at least one page on the internet that mentions Claude as an AI? Well, then it's part of the training material.


Yes, that's why Claude might think it's ChatGPT from time to time, because it scraped a url that contained chatgpt conversations.

But the reason why Kimi thinks it is Claude is different, it's because chinese companies make thousands of fake accounts, fire prompts at Claude and train their models to imitate Claude.


That’s how a smoking gone turns into a water pistol


when open source models are banned maybe i will become a cartel kingpin smuggling open source weights into the usa. find a sufficiently shifty street corner, 'what do you need', "kimi". are you familiar with my product? pure mxfp4, $500 per TB.

(joke and walter mitty, i know the government reads my messages)


if this qualifies bootleg, point me at non-bootleg frontier ai.


Clearly, the best system is the one where you spend the most to get less /s


Our customers across EU (hospitals) are not impressed or interested (n=175). Such a delusional project.

The ICC move by MS made hospitals go in an even higher gear to prepare off-ramp plans. From private Azure cloud to “let’s get out”


My friend Alexander created an easy docker file[1] for an OSM tile server. I use it for nearly two years without problems.

[1] https://github.com/Overv/openstreetmap-tile-server


Joan’s parent company, Visionect[1], also produces the Place & Play, a 13” or 36” eInk display[2]. Some guy created a rotating newspaper frame from it[3]. Looks quite amazing, and it has a 1 year battery life.

[1] https://www.visionect.com

[2] https://www.visionect.com/product/place-and-play-32/

[3] blog: https://gregraiz.com/i-made-an-eink-newspaper/ and video: https://youtu.be/gECj1AE9D2c and code: https://github.com/graiz/newsprint


There has been quite some research on hyperbaric oxygen therapy and other neurological diseases.

My dad was involved in MS research in the 80s utilizing hyperbaric oxygen. After some hopeful results, no clear evidence was found.


I moved with my girlfriend from our apartment in Amsterdam to her family holiday home on an island in South-Eastern Europe. We save a decent amount of cash (our rent+utilities was 2000€ alone) and live an amazing life here.

Definitely considering to find remote-only positions in the future. Currently my company allows me to wait out COVID anywhere within roughly the same timezone.


This doesn't count unless girlfriend's parents sleep in an adjoining bedroom.



Django abstracts away a bit too much, if you ask me. For a simple REST API it's complete overkill. I prefer Flask or Falcon[0], in which you don't just get better performance[1] but also a lower level of control.

[0] https://falconframework.org/

[1] https://klen.github.io/py-frameworks-bench/ or https://github.com/the-benchmarker/web-frameworks


Why exactly do you want a lower level of control?

You want the complete opposite; have as many problems solved for you already with battle-tested ways to break out of the happy path when you need to.

Every sufficiently large Flask app becomes a half-assed reinvention of half of Django, except without the hundreds of thousands of eyeballs and man-centuries of bug fixing.


I agree with your parent. When I started with (Python) webdev some years ago I picked Flask, but nowadays I use Django, as it has superb documentation and useful things built-in. In particular I miss the user-management, authentication and authorization in Flask.

Personally I consider Django without any additional apps less complex than Flask with apps (or "plugins" in Flask jargon) to handle basic stuff like auth.

If performance is a concern, both of them are insufficient and I would rather pick another language like Rust, Go or even Java.


I started with Flask, moved to Django, and now at work am back on Flask. It all depends on what you are building; if you are going to need some ORM, standard auth, etc, Django is a no brainer. However if you are going to be building out much of your own idiosyncratic architecture, Flask gets out of your way and avoids including lots of things you don't need.

For performance, we are building some parts of our system with Clojure. It feels extremely well suited, albeit with some initial learning curve.


How are you implementing REST using Faclon if I might ask ? fully custom ? or is there anything like DRF hanging around ?


I am experimenting with GPS luggage tracking hardware with simple live-updates. This is potentially a great use-case.


That's cool! How do you deal with tracking while it doesn't have GPS signal?

WiFi I suppose, using a mapping from BSSID -> lat/long, like Mozilla Location Services or Combain.

What's the power consumption like from attempting to get a GPS lock? Luggage presumably spends most of it's life indoors. I imagine a WiFi scan uses less power than attempting to acquire a GPS signal.

I'm a hobbyist when it comes to hardware/IOT, I've been working on something similar but for RV's instead of luggage. Much harder constraints when it comes to luggage! RVs have power and room to spare.

Oh, you could put a small generator on the luggage wheels, to charge it a bit while wheeling the luggage around!

Sounds like a fun project.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: