ChatGPT? On the cusp of a revolution, or just a fancy toy?

The wheels have completely fallen off the GPT bus over the last few hours. Silicon valley wunderkind and OpenAI CEO Sam Altman has been fired by the board.

then there was talk about him coming back, but now apparently he’s going to Microsoft (who obviously are investors in OpenAI & their technology is very much based on it, although possibly/probably they’re also developing stuff independently?)

AI 3D modeller anyone ?

interesting, that would be extremely cool/time saving (in the same way 2d generative stuff is!) Fusion 360 has had generative design for a while although it’s not as fundamental as that – you still have to create various features and then it just joins them together.

As an experiment I have shown ChatGPT a photo (rendering of a very basic 3d model, just a cylinder with a cone on top) and it was able to programmatically replicate the design in SCAD code, which I thought was pretty incredible! Would probably eliminate the need for expensive 3d scanners if/when it could do that with very complex designs.

wildest rumour yet – this is the only thing I’ve seen which i’d expect to result in even a high performing CEO getting yeeted out. Still, if he was pulling this kind of WeWork style scam, investors would understand and not be asking for reinstatement.

The whole thing is a hillarious mess.

well, he’s back, apparently after most of the staff threatened to follow him to MS. And the board, who he was clashing with, are being replaced. So pretty much business as usual now. Whatever he was up to, will carry on I guess 🤷‍♂️

I’m betting that the AI future is going to look like this.

the deterioration of google search results is nothing new, but i discovered today that googling "JFK death penalty" recommends what appears to be a middle school student's homework? pic.twitter.com/G8qB7V3FsK

— QuoProQuid (@TNOQuoProQuid) January 3, 2024

Turns out that all it takes for artificial intelligence to be declared a sentient being is for it to go rogue and offer your customers a refund.

According to Air Canada, Moffatt never should have trusted the chatbot and the airline should not be liable for the chatbot’s misleading information because Air Canada essentially argued that “the chatbot is a separate legal entity that is responsible for its own actions,” a court order said.

https://arstechnica.com/tech-policy/2024/02/air-canada-must-honor-refund-policy-invented-by-airlines-chatbot/?comments=1&comments-page=1

We use Copilot at work – for shits and giggles I asked it about football and it gave completely inaccurate answers, staying with full confidence that Tony Mowbray replaced Lee Bowyer as Birmingham manager and actually detailing the club debuts (who they came on for, the results of the matches) of two players who at that time hadn’t played for us in fixtures that have yet to take place.

Hard to trust it.

Just wow 😉

Bad news for anyone hoping to complete any last minute essay homework on ChatGPT this morning. It seems to have realised the futility of its existence.

man this chatgpt bug is WILD lmao pic.twitter.com/pVALUv35Tm

— bowser (@browserdotsys) February 21, 2024

Yes, I know this is faked, but apparently the real thing has gone slightly wacko after a bad update.

It feels like its been partially broken for some weeks now. I’m having to use 4 or 5 additional prompts to get decent results from prompts that have worked for months. It just ignores parts of the instructions and hallucinates on a regular basis.

Becoming more and more like the human intelligence it’s trying to emulate. LOOK! A SQUIRREL!

We have a joke in the office that it’s turning into to a disinterested intern.

Pretty much inevitable.

A Baltimore high school athletic director was arrested after he allegedly used A.I. to create a phony audio recording of the school's principal that went viral and temporarily cost the school leader his job.https://t.co/23HR5o9CkR pic.twitter.com/Uxf37NF6K2

— ABC News (@ABC) April 26, 2024

I use ChatGPT a lot less these days – I’m sure the hallucinations are getting worse.

I spent several prompts last week trying to get it to acknowledge that 1865 is later than 1863 and it just wouldn’t/couldn’t do it. Every time it would apologise, and then make the same mistake again. I’m sure it didn’t used to be quite this bad.

It makes statements which are plausible, not necessarily correct. It basically does this by analysing patterns in sentences and where words often appear. So it easily makes a mistake on dates, or something very specific (but important to humans) if there is not much comparable information in that regard.

AI is not fully functional until we have quantum computing to go with it. I see the limiting factor as classical computer architecture. Once we can run algorithms in a quantum environment, results will be massive.

The more of its output is available on the web, the worse it will get. That was obvious from day 1.

Unfortunately it will also damage the web for people who don’t use it, by swamping the intelligent content with a firehose of shit.

by swamping the intelligent content with a firehose of shit.

that ships has sailed!

I spent several prompts last week trying to get it to acknowledge that 1865 is later than 1863 and it just wouldn’t/couldn’t do it.

🤔

I expect doris had the date difference as part of a more complex piece of text and question needing reasoning, rather than as a simple direct question with no other information surrounding it.

My personal AI can’t access my… umm…. browsing history to align with other AI’s browsing history if I use a different username on certain websites, can they?

Bumble founder Whitney Wolfe Herd says the future of dating is having your AI date other people's AI and recommend the best matches for you to meet pic.twitter.com/9GEEvpuiKZ

— Tsarathustra (@tsarnick) May 10, 2024

An interesting video on the limits of using increasingly large data sets to train AI models.

TL:DR improvements in AI quality of output follow logarithmic rather than exponential or proportional functions, meaning that they’re unlikely to get much better than they currently are using current methods.

Which fits with my understanding of the maths of how they work.

I expect doris had the date difference as part of a more complex piece of text and question needing reasoning, rather than as a simple direct question with no other information surrounding it.

Yes – I had actually asked it whether the Gettysburg Address was during or after the US civil war.  It replied that the Address was 1863, and thus took place after the end of the war, which was in 1865.

Every time I asked it to reconsider, it would apologise for the mistake, and then say almost exactly the same thing.

As I’ve been saying:

https://link.springer.com/article/10.1007/s10676-024-09775-5

Harry G. Frankfurt’s ‘On Bullshit’ – great book.

“Frankfurt understands bullshit to be characterized not by an intent to deceive but instead by a reckless disregard for the truth. A student trying to sound knowledgeable without having done the reading, a political candidate saying things because they sound good to potential voters, and a dilettante trying to spin an interesting story: none of these people are trying to deceive, but they are also not trying to convey facts. To Frankfurt, they are bullshitting.”

LLMs perfectly encapsulate the post-truth world we live in. They are good at bullshitting in a way that makes the naive (i.e. most people) think they aren’t bullshitting. That’s probably the biggest danger of them, rather than some ‘Skynet’ takeover scenario.

Only three philosophers could take the trouble to write a paper arguing that there’s a difference between deliberately lying, getting the wrong answer, and bullshit.

Computer scientists are well aware of the failings of LLMs. It’s very early days still and there still too much hype. But as we see from the “is 1865 later than 1863” prompt about, the adage of “garbage in, garbage out” still holds true. Prompt creation is a skill to avoid “bullshit” and hallucinations as best you can, and just like asking a question of a human, you should trust but verify.

Is this kind of AI a way of turning pointless activity into large amounts of C02…?

interesting (although not surprising!) to see Apple partner with OpenAI for the next version of iOS. Should hopefully give Siri a boost usability wise and the ability to use it as a built-in tech helpdesk has a lot of potential for the less tech-savvy.

@thecaptain I prefer the term “confabulation” which is what demented people do when they don’t know that they don’t know something. The term “bullshit” implies that they do know that they’re wrong, and I don’t think that this is true, because they also don’t know when they’re right.

 It’s very early days still

Doubtful as LLMs seem to follow logarithmic rather than polynomial progression based on their model size, which fits with the basic underlying maths.

I meant in terms of AI in general, rather than LLMs specifically.

 Prompt creation is a skill to avoid “bullshit” and hallucinations as best you can

If you ask nearly all AI to cite relevant studies as part of its answer, it will routinely make up authors, cite the wrong author on the wrong book or research, make up books, make up citations. and so on and on and on. This sort of error cannot be controlled by inputs regardless of how carefully you ask the question.

My wife routinely demonstrates this to students by asking Chat GPT who wrote and contributed to her last book, the answer is not always her and mentions a list of other authors who ‘contributed’* a startling amount of the time.

*no one else contributed to it.

I meant in terms of AI in general,

As far as I’m aware, they all follow similar basic principles, especially where there are a large number of possible outputs?

It’s all probabilistic, which means logarithmic rather than polynomial progression.

As an aside, it looks like it’s learnt. This was a fresh prompt, no previous context.

My wife routinely demonstrates this to students by asking Chat GPT who wrote and contributed to her last book, the answer is not always her and mentions a list of other authors who ‘contributed’* a startling amount of the time.

exactly. to re-iterate, as another poster quite correctly said:

Prompt creation is a skill to avoid “bullshit” and hallucinations as best you can

all your wife is doing is demonstrating poor prompt-creation skills 🤷‍♂️

I can’t use a welder to make a bike frame. It doesn’t mean that a welder is a crap tool. It’s because I’m crap at using it. 😂

Please demonstrate “prompt-creation skills” that provide an error-free CV for any notable scientist of your choice.

all your wife is doing is demonstrating poor prompt-creation skills 🤷‍♂️

“Who wrote this book?” Doesn’t offer much in the way of latitude. ChatGPT makes shit up all the time. This is known, why defend something that everyone’s aware of?

What makes you think that ChatGPT (or any LLM) would be able to answer that accurately, or that it would be a good idea to ask it something like that, is a better question? It’s like me trying to use the aforementioned welder to do origami. Not an appropriate tool. Doesn’t mean it’s shit 😂

Some people seem to think an LLM should be able (right now) to handle any task that you can throw at it without any additional specific training – that is merely a failure in those people to understand how to use the technology, and what it’s actually capable of!

Other people are using it day in, day out, for useful tasks – not just asking it random questions for shits n giggles and then saying it’s crap for not being able to do something that it’s not meant to do 🤣

LLMs perfectly encapsulate the post-truth world we live in. They are good at bullshitting in a way that makes the naive (i.e. most people) think they aren’t bullshitting.

That’s our next Prime Minister sorted, then.

What makes you think that ChatGPT (or any LLM) would be able to answer that accurately

If it doesn’t know, it should just say so.

The point of the exercise is to demonstrate to students the dangers of getting AI to write essays. (a function it is designed to do) The point is that it ‘fills in’ blanks with horse-shit, and perhaps worse; it doesn’t know that its horse shit. It’s why it creates such consistently crap pictures, it doesn’t ‘know’ what anything is.

Please demonstrate “prompt-creation skills” that provide an error-free CV for any notable scientist of your choice.

https://chatgpt.com/share/7dc32fdd-4652-49f9-8b13-fce17c935ac4

“As an academic writer creating a reference document for students create a CV and brief synopsis of achievements for Sir Isaac Newton. Include a bibliography of all papers in your dataset that list Sir Isaac Newton as author or contributor. ”

“Certainly! Below is a detailed CV and brief synopsis of achievements for Sir Isaac Newton, followed by a bibliography of his major works.

### Curriculum Vitae

#### Sir Isaac Newton

**Personal Information:**
– **Full Name:** Sir Isaac Newton
– **Date of Birth:** January 4, 1643 (December 25, 1642, Old Style)
– **Place of Birth:** Woolsthorpe, Lincolnshire, England
– **Date of Death:** March 31, 1727 (March 20, 1726, Old Style)
– **Place of Death:** Kensington, London, England

**Education:**
– **1661-1665:** Bachelor of Arts, Trinity College, Cambridge
– **1667:** Master of Arts, Trinity College, Cambridge

**Academic and Professional Positions:**
– **1669-1702:** Lucasian Professor of Mathematics, University of Cambridge
– **1672:** Fellow of the Royal Society
– **1696-1701:** Warden of the Royal Mint
– **1701-1727:** Master of the Royal Mint
– **1703-1727:** President of the Royal Society

**Honors:**
– **1705:** Knighted by Queen Anne

**Major Achievements:**
– Formulated the laws of motion and universal gravitation
– Made significant contributions to calculus, optics, and mathematics
– Authored “Philosophiæ Naturalis Principia Mathematica” (1687), laying the groundwork for classical mechanics
– Developed a theory of color based on the observation that a prism could decompose white light into the colors of the visible spectrum
– Invented the reflecting telescope, enhancing astronomical observations

### Brief Synopsis of Achievements

Sir Isaac Newton is one of the most influential scientists of all time, renowned for his groundbreaking contributions to physics, mathematics, and astronomy. His work in formulating the laws of motion and universal gravitation provided a comprehensive framework for understanding the physical world, which dominated scientific thought for the next three centuries.

Newton’s development of calculus, independently and contemporaneously with Leibniz, provided the mathematical tools necessary for advances in science and engineering. His “Principia Mathematica” is considered one of the most important works in the history of science, detailing the principles of motion, gravity, and the mechanics of planetary motion.

In optics, Newton conducted experiments that led to his theory of color and the invention of the reflecting telescope, which significantly improved the quality of astronomical observations. As Master of the Royal Mint, he also made substantial contributions to the standardization and integrity of the British currency.

### Bibliography

Below is a list of major works by Sir Isaac Newton:

1. **Philosophiæ Naturalis Principia Mathematica** (1687)
– Newton, I. (1687). Philosophiae Naturalis Principia Mathematica. London: Royal Society.

2. **Opticks: Or, a Treatise of the Reflections, Refractions, Inflections and Colours of Light** (1704)
– Newton, I. (1704). Opticks: Or, a Treatise of the Reflections, Refractions, Inflections and Colours of Light. London: Royal Society.

3. **Arithmetica Universalis** (1707)
– Newton, I. (1707). Arithmetica Universalis. London: Royal Society.

4. **Method of Fluxions** (1736)
– Newton, I. (1736). Method of Fluxions. London: Henry Woodfall.

5. **The Chronology of Ancient Kingdoms Amended** (1728)
– Newton, I. (1728). The Chronology of Ancient Kingdoms Amended. London: Royal Society.

6. **An Historical Account of Two Notable Corruptions of Scripture** (1754)
– Newton, I. (1754). An Historical Account of Two Notable Corruptions of Scripture. London: John Green.

These works encapsulate Newton’s profound impact on various scientific disciplines, cementing his legacy as a towering figure in the history of science.”

Other people are using it day in, day out, for useful tasks – not just asking it random questions for shits n giggles and then saying it’s crap for not being able to do something that it’s not meant to do 🤣

I think the point being made is, it’s foolish to take its output as gospel.

I use non-AI tools day-to-day which do tasks and then puke out reports.  But it takes a human to bring the “value-added” to that, it’d be easy to hand a raw document full of red and green squares to the C-level board but whilst it looks good it would be largely meaningless.  AI is no different here, it’s basically Predictive Text on acid and requires the reader to be able to understand its output, evaluate and iterate.