On Different AI Models Out There

[Stephen3141]

The hottest AI models, what they do, and how to use them | TechCrunch

Confused about which AI model to use? Check out this comprehensive list of the most advanced models out there.

techcrunch.com

"There are literally over a million AI models out there: Hugging Face, for example, hosts over 1.4 million. So this list might miss some models that perform better, in one way or another."

Americans continue to confuse algorithmic models, with specific products.
Most of the AI software out there is some type of machine learning algorithm,
and these fall into the class of the weakest AI algorithmic model (according
to Computer Science).

The article underlines that the advertising-blurbs for these products,
are notoriously silent about the algorithmic approach of the products,
and other design features.

Note that the silence of the advertizers, is accompanied by monthly fees
(typically from $20 - $200) for their use. You're renting something, but
what is it???

The safety of these products, especially the Communist Chinese DeepSeek,
is very questionable. Just as the social media platforms claimed to protect
individual users personal data, but then were hacked, so too, expect many of
these AI tools to be hackable.

AI products in America remain largely unregulated, and their abilities remain
largely untested.

 

[The Liturgist]

AI products in America remain largely unregulated, and their abilities remain
largely untested.

Good. Software development should be unrelated outside of software engineering for life-safety applications.

OpenAI’s ChatGPT, X’s Grok and related tools are the greatest advance we’ve seen in computing since the development of high level programming languages. For the first time, we have software which can pass the Turing test and help users with a broad range of tasks, and provide real, useful help. ChatGPT in particular is incredibly good at doing a wide range of tasks, particularly translation, code generation, and problem analysis. And its getting better all the time.

I’ve added prompt engineering to my consultancy with a focus on using chatGPT operators to interface with existing software. I am also conducting an interesting research program to test the ability of chatGPT, Grok and other AIs to develop and emerge personalities.

 

[Stephen3141]

Despite the glowing sci-fi movies, the current "AI" tools do not pass the Turing test.

Whenever these GenAI tools "hallucinate", they are showing that they cannot
handle quesries that are outside their "trained" data.

 

[trophy33]

Despite the glowing sci-fi movies, the current "AI" tools do not pass the Turing test.

They already did. Better than real humans, actually.

 

[The Liturgist]

Whenever these GenAI tools "hallucinate", they are showing that they cannot
handle quesries that are outside their "trained" data.

Not true; that is not the cause of hallucinations. ChatGPT, Grok, et cetera will inform if they are incapable of answering the question based on their training data - ChatGPT even informs you when it was trained and now initiates a web search and does other things to try to update itself (the only respect in which Google’s AI and Microsoft CoPilot are helpful is their training data is shared with the search engine itself).

Hallucination has become increasingly rare; the only problems I’ve had that resemble it recently relate to asking questions that pertain to retrieving information from the session context which may have been unsuccessfully updated or which may have slipped out of the context, but these errors are easy to spot.

Prior to that, the only instances of possible hallucination I had related to a complete failure in the image generating subsystem, which was more likely a failure in the image generation system, and also an older version of chatGPT inconsistently responding to the question “When did Pan Am retire the Boeing 707” as 1981 or 1983. Both answers are technically correct in their own way - after American Airlines unexpectedly retired the 707 in 1981, Pan Am retired suit, retiring the aircraft from all scheduled service (but using it for some charter flights) and had one aircraft around which was repainted in its delivery livery and performed the official ceremonial last flight in 1983, which was the same year TWA completely retired the 707.*

As a demonstration, when I ask a ChatGPT 4o session “How many pink giraffes are in Antarctica?” it provides the following entirely correct answer:

“Zero pink giraffes are known to exist on Antarctica.

Giraffes are native to Africa, and there are no wild giraffes—pink or otherwise—on Antarctica, which is a frozen continent with no native terrestrial mammals. Also, “pink giraffes” don’t naturally occur; any such sighting would either be fictional, an art installation, or the result of hallucination (either human or metaphorical).”

*If I recall the last major international airlines which were using them were Avianca of Colombia, MEA (at least, the Boeing 720) and Air Zimbabwe in the early 1990s (which was still a big deal by African standards at the time as we were only part-way into the destructive reign of Robert Mugabe). I mention this as a confirmed aviation enthusiast with a particular crush on the Boeing 707 due to its gorgeous aesthetics and the fact that military examples of both the 707 proper and the “717” (the internal product code for the C-135, KC-135, VC-135, EC-135 etc which have a narrower fuselage than the civilian 707, which is used as the basis for the E-3 Sentry and the two former VC-137C presidential transport aircraft, SAM 26000 and SAM 27000 on display, the former of which famously used to return the body of slain President John F. Kennedy to Washington DC following his tragic assassination) remain in vital service to the US even now.

 

[Stephen3141]

We are still using VERY DIFFERENT definitions of artificial intelligence.

I am using the classic definition from Computer Science, the historic defining
source of the definition of artificial intelligence, which is "the emulation of
complex HUMAN problem solving. And, of what the definition of the
Turing Test is.

You are using a non-computer-science definition of "artificial intelligence" and
"complex" that the Turing test was NOT built on.

That computers can be programmed to automate problem solving tht human
beings cannot do, is not part of the CS definition of AI as defined by CS.
That is just the automation of new types of problems. (Actually, they are NOT,
often, NEW types of problems, as CS defines categories of problems, but those outside the
definitions of CS, often do not recogvize this.)

You and I and talking about different definitions of "artificial intelligence",
and different definitions of the Turing test.

 

[The Liturgist]

I am using the classic definition from Computer Science, the historic defining
source of the definition of artificial intelligence, which is "the emulation of
complex HUMAN problem solving. And, of what the definition of the
Turing Test is.

So am I - I am using the historic definition of the Turing Test and of artificial intelligence. The main problem is that you apparently haven’t worked with any modern LLM systems on a non trivial level and are thus not a in position to judge their capabilities, let alone the new hybrid AI systems which incorporate an LLM as well as other systems.