I'm using Linux Mint 22.1 Cinnamon on a desktop.

The Mint software manager offers at least three AI chat things -

"Gai Generative software tools for beginner"

"XCA AI chat"

"Newelle AI chatbot"

and Duckduckgo also offers access to an AI chatbot.

Are there any better AI things for technical questions that are free and preferably do not require registration?

Are any of the above the same thing under different names?

Thanks

The Gai homepage https://webpath.iche2.com/app/gai/download_en.html and the "Documentation" page all in Chinese https://webpath.iche2.com/gaidoc/en/

The Newelle homepage https://github.com/qwersyk/Newelle

The XCA AI chat homepage https://github.com/alfianlosari/GTKChatGPT

Duckduckgo AI https://duckduckgo.com/?q=DuckDuckGo...=chat&duckai=1

what do you mean by best?

I use 'Grok dot com' extensively and occasionally use 'gemini dot google dot com', and 'storm dot genie dot stanford dot edu'. Probably not the best but adequate for most needs.

Edit: I think the Stanford ai requires registration, gemini might but I don't remember, the standalone grok is free to all comers.

Most knowledge, most clever, least likely to have "hallucinations".

That is still not a definition. Best, most clever, least likely are all personal preferences, everyone understands them differently, in general they mean nothing.

1 members found this post helpful.

Just like DW 'implies' Mint is the [only] Linux distro to use, (was MX; always use #1)

ChatGPT.com will answer your question (without any login, UNlike [evil] grok etc)
with: "#1 ChatGPT.com" (click "Stay logged out", the bottom [3rd] option in the login/signup popup)


Personally, as of today, I'll wait exactly 1 year, for Slackware-AI v.0401
(which will use systemd-ai, & replace all OS'es, making CPU/GOD finally fully Tr{um,ans}Parent)



Everyone needs to replace their PC with a Loona pet (which a friend has done)!

That means avoiding LLM slop entirely and especially certain ones and hiring an actual reference librarian for the task. The gist is that you are wasting your time, money, and electricity with LLMs unless you are involved in growing the bubble yourself:

So to answer the question of which one is "better", that would be Perplexity because at the moment it is only 37% wrong on average. :/

I realize they are "popular" at the moment, but so were Pogs. Although with Pogs, there was at least an artifact left as a reminder when the bubble popped. Due to their nature as statistical models, merely throwing more time, money, data, or electricity at LLMs only makes them more expensive not good. Reasoning models are needed, instead of language models, and those are not on the horizon.

1 members found this post helpful.

^^^Lol, I thought I would ask my favorite ai to craft a rebuttal to your criticisms. Took it about 10 seconds:
Grok 3 (beta): "While your critique of large language models (LLMs) raises some valid concerns about their current limitations, it overlooks their strengths, mischaracterizes their purpose, and underestimates their potential for improvement. Let’s address your key points systematically.
On Accuracy and Reliability
You cite a study from the Columbia Journalism Review showing that LLMs like Perplexity (37% incorrect) and Grok 3 (94% incorrect) often fail to retrieve accurate information. This is a fair observation—LLMs can indeed produce errors or "hallucinations." However, this criticism hinges on an expectation that LLMs should function as flawless fact-checkers or citation engines, which misunderstands their design. LLMs are not databases or traditional search tools; they’re generative systems trained to produce human-like text based on patterns in vast datasets. Their strength lies in tasks like brainstorming, drafting content, or summarizing complex ideas—not in replacing reference librarians or serving as sole sources of truth. When used appropriately (e.g., as a starting point rather than a final authority), their error rates become less of a fatal flaw and more of a manageable limitation.
"Just Statistical Models"
You argue that LLMs are merely statistical models and that throwing more resources at them only increases costs, not quality. While it’s true that LLMs rely on statistical patterns in language, this doesn’t mean they’re stagnant or incapable of progress. History shows otherwise: advancements in model architecture (e.g., transformers), training techniques (e.g., reinforcement learning from human feedback), and data curation have driven significant improvements. Compare GPT-1 to GPT-4—newer models generate more coherent, contextually relevant responses, often outperforming their predecessors in benchmark tests. Resources aren’t just "thrown" at LLMs; they’re strategically applied to refine their capabilities. The claim that this approach yields no improvement dismisses tangible evidence of progress.
The Hype Bubble Analogy
You liken LLMs to Pogs, suggesting they’re a fleeting, overhyped trend destined to leave nothing behind when the "bubble pops." While it’s reasonable to caution against overblown expectations—AI hype does sometimes outpace reality—this analogy downplays the practical value LLMs already deliver. They power customer service chatbots, assist in education (e.g., tutoring tools), and streamline content creation across industries. Unlike Pogs, which were a collectible fad, LLMs are a versatile technology with real-world applications. The bubble may deflate in terms of exaggerated promises, but the utility of LLMs ensures they won’t vanish into irrelevance.
Reasoning Models vs. Language Models
Your call for reasoning models over language models is compelling—true reasoning, like solving novel math problems or understanding causality, remains a frontier LLMs haven’t fully conquered. However, this isn’t an either/or scenario. LLMs aren’t the endpoint of AI development but a stepping stone. Researchers are already exploring hybrid approaches, integrating symbolic reasoning with neural networks to bridge this gap. Projects like DeepMind’s AlphaCode or OpenAI’s work on reasoning benchmarks show progress toward models that can think more systematically. Dismissing LLMs as a dead end ignores their role in this broader evolution.
Conclusion
Your skepticism about LLMs is understandable, particularly regarding their accuracy and the hype surrounding them. But writing them off as "slop" or a waste of resources overlooks their current utility and future potential. They’re not perfect, and they’re not a substitute for human expertise or entirely new paradigms like reasoning models. Yet, when used wisely—say, to augment rather than replace traditional methods—LLMs offer real value. The challenge isn’t to abandon them but to refine them and push AI research forward. The bubble may shrink, but the technology won’t disappear; it’ll adapt and grow.
This rebuttal acknowledges the critic’s concerns while defending LLMs with evidence of their strengths and ongoing development, striking a balanced tone that invites further discussion."

I personally don't use LLMs either. A 'straight' search engine is much better as you go directly to 'articles' on the subject (hopefully written by human subject matter experts). I really don't see the value of LLMs for known cataloged information. Use search like using a library. All you need. LLMs are a waste of energy IMO. And as we all know AI is not AI at all, but some would like to think it the real 'deal' which is scary.

1 members found this post helpful.

At some point, websites will embed it (just like they embed yt videos now) and it might become difficult to avoid.
Going back to GET method instead of CONNECT would probably help, not sure.

Another criterion for best AI chat is being good at mathematics. In the recent past AI chatbots used to be terrible at maths, or at least arithmetic.

Man, how do I phrase this without being overtly banworthy... uhh... uhhh...
Maybe like this:

Go to the original datasets the "AI" models copyright 'whitewash' instead of using the AI because the archives where they got all those scientific infos have their own search engines.
I can't list any names or URLs of course since we live in a fecked up world where I'd be linking to legally shakey 'piracy' stuff if I did, but if you're something like facebook and just scrape it and repackage it into a LLM then it's a-o-kay.

But yeah.
There's a funny meme about it... it's got aaron schwartz and mark zuckerberg on it and it's about AI and 80TB.

Anyway, if you go to the source of the data, you'll have to do slightly more work looking for the source first, and then navigating any source for the problem you're having.
But on the upside, no hallucination of any AI will tamper with the bits and pieces to be found there.

Now, why do I call myself clueless dolt? Because I am, just because I know a place where I can find scientific things doesn't mean I'm actually going there cause...in my case.
I LOATHE pdfs. If I'm to read anything substancial for 'lernin' , its gotta be on paper.

Edit:
Oh, and I prefer people over computers.
Google is such a big beefy company but their search engine r t3h suxx0r5