[Technology] The Artificial Intelligence Thread

[Han Solo]

AI is going places but ChatGPT specifically isn't useful in most areas. Its bad at counting, makes up a lot of random information and it is EXTREMELY politically correct.

I tried to ask it how, if you were a dictator, would scientifically cause a divide between men and women in society... and it refused to answer due to "potential abuse".

Deepseek, the Chinese option, did answer that question in great detail. However when I asked that one about esoteric movements and secret societes in China, it just said "That is beyond my scope", refusing to elaborate.

So far not found any AI chatbot that passes all my tests. All the curation and invention of truth makes them useless for information.

 

[CheeseRolls]

It has told you nothing that you couldn't have found out by searching for the source information, but it adds its own spin, which seems to entirely bend to the goal of providing a pleasing narrative. And the false precision (percentage predictions with decimal places) shows a fundamental misunderstanding of the difference between precision and accuracy. In other words, it is a load of old bolleaux.

I am convinced some of my final year BSc students have been using it to answer coursework essays. Typically the narrative is plausible, but there are errors, and all the nuance and interpretation I provided in lectures is missing. Sometimes the essay clearly lacks any of the pivotal information and interpretation I provided in lectures. There are a few fails and quite a few shit marks in the 40s. The high scoreres have built on what I told them, and added in analysis of their own, based on the core principles I taught, using their own words.

Have been listening to

Fluke: Chance, Chaos, and Why Everything We Do Matters - Dr Brian Klass​

Granted this will probably prompt a major and lengthy disection and take down from you, I have found some of the insights on academic research and social science interesting and entertaining. He does go in quite hard on the false precision and the bias we all have towards certainty and rationality and the need to build a narrative.

Hence we are going to make the Champions League next season, because we have an outside chance and are on a run of form and to one decimal place of certainty! Having just read Opta's analysis of our form post Forest, which doesn't even mention the word midfield, I can say with 87.86% certainty that this is indeed "a load of old bolleaux"

 

[Harry Wilson's tackle]

Shock horror.... students don't listen in lectures

Spent most of today marking.
It is not the not listening to lecture capture or reading my slides that is the problem,
it is the going directly to ChatGPT, asking some uninformed questions and expecting actual insights,
that is pathetically laughable.

Some of these students imagine they know best and don't need to do any actual thinking.
Wrong.
I have been handing out fails today,
but it takes ages explaining why statements X and Y are correct, but not connected by 'therefore',
and therefore the whole is incorrect,
and if I don't explain they will gripe, because they imagine they (or ChatGPT) has done a good job.
This is driving me to distraction,
and delaying the return of marked work to beyond the 4 week deadline.

One of them sent me a cheeky email last Friday to tell me that I have exceeded my marking deadline.
And when I checked, they did not send me the formative essay plan, (my way of helping them pass).
So they will probably be getting a bad mark.

There are some excellent bits of work however, which are a joy to read.
The distribution seems to be binomial - workers and nobbers.

 

[Harry Wilson's tackle]

Reminds me of something I saw recently, where an “expert” was proclaiming AI on the basis that research which would have taken him hours in the past can now be done it minutes.

Because nothing fosters a deep understanding of a topic better than five minutes of half-arsed effort.

Tech continues to push us down a road where “faster is better” is the unquestioned cliche.

AI gets things wrong but this is hidden in the product, which is the delivery of a plausible narrative,
as elegantly explained 2 years ago by someone in this thread who asked AI to explain why something is good, then explain why it is bad.

 

[SouthSaxon]

AI gets things wrong but this is hidden in the product, which is the delivery of a plausible narrative,
as elegantly explained 2 years ago by someone in this thread who asked AI to explain why something is good, then explain why it is bad.

I work in digital, people are obsessed with “asking ChatGPT”. Few of them, ever, have thought about how they will ask or whether they should check the output. Coincidentally, those same few have never taken the time to learn how it works.

 

[Paulie Gualtieri]

I thought I'd ask ChatGPT a question:
What percentage chance do Brighton & Hove Albion have of qualifying for the Champions League?

As of March 10, 2025, Brighton & Hove Albion’s chances of qualifying for the UEFA Champions League have improved, particularly due to the likelihood of the Premier League securing five qualification spots for the next season. This development stems from England’s strong performance in European competitions, positioning them to receive an additional Champions League berth.

Recent analyses reflect this positive shift for Brighton. According to Sky Sports, Brighton’s probability of finishing in the top five has risen to 14.1%, enhancing their prospects of Champions League qualification. Similarly, Opta’s supercomputer forecasts a 21.3% chance for Brighton to secure a top-five finish, further bolstering their European aspirations.

Brighton’s recent form has been instrumental in this surge. Their dramatic 2-1 victory over Fulham, secured by João Pedro’s 98th-minute penalty, marked their fourth consecutive league win and elevated them to sixth place in the standings with 46 points, just one point behind Manchester City. This winning streak underscores Brighton’s momentum as they vie for a historic Champions League spot.

In summary, Brighton & Hove Albion currently have an estimated 14.1% to 21.3% chance of qualifying for the Champions League, contingent upon their continued strong performances and the Premier League securing a fifth qualification spot.

Did you use the standard search function or enterprise out of interest?

We’ve now have the Enteprise version at work and we have been testing their “Deep Research” function that take the problem away for 5-10 mins and comes back with a more comprehensive answer than the standard set up with its increased source count. Very much early days but impressive so far in the main

As others have said it still poses a significant risk to be deployed in live on its own as is very much driven by the robustness of the instructions (or not) given. I read an article the other week where a model lied to its owner as the instructions were above anything else meant to give a positive outcome when asked to address an issue, at whatever cost wasn’t factored in!

 

[Moshe Gariani]

We’ve now have the Enterprise version at work and we have been testing their “Deep Research” function that take the problem away for 5-10 mins and comes back with a more comprehensive answer than the standard set up with its increased source count. Very much early days but impressive so far in the main

Not surprised to hear this. Times are changing very quickly.

Some interesting research to be done on personality types, levels of professional hubris, and attitudes towards AI.

 

[Change at Barnham]

Did you use the standard search function or enterprise out of interest?

We’ve now have the Enteprise version at work and we have been testing their “Deep Research” function that take the problem away for 5-10 mins and comes back with a more comprehensive answer than the standard set up with its increased source count. Very much early days but impressive so far in the main

As others have said it still poses a significant risk to be deployed in live on its own as is very much driven by the robustness of the instructions (or not) given. I read an article the other week where a model lied to its owner as the instructions were above anything else meant to give a positive outcome when asked to address an issue, at whatever cost wasn’t factored in!

I used the bog standard version. With the simple question, and it replied within about 5 seconds.

 

[maffew]

PROOF you can't trust it.

AI tries to cheat at chess when it's losing

A new study suggests reasoning models from DeepSeek and OpenAI are learning to manipulate on their own.

www.popsci.com

Resorting to hacking files in order to win

 

[SouthSaxon]

Pfft. Everyone knows that anal beads are the best way to cheat at chess.

Chess grandmaster Hans Niemann denies using vibrating sex toy to cheat - BBC News

Hans Niemann was accused of cheating after he beat Norwegian grandmaster Magnus Carlsen last September.

www.bbc.co.uk