The Anything Thread that is Never off subject.

[Woosh]

What if US does sanction the UK censors..
Then the EUSSR censors..
No US travel, no Visa, no Mastercard, frozen accounts, brilliant!
Some of their own medicine.

If you value protection from harm and social cohesion, you might see UK's and EU's anti hate speech laws as justified and necessary.
So where do you stand?

I am not surprised that some people think that free speech should be totally, absolutely free.
Imagine a world you will be arrested based on someone else's opinion.
Next time you go to Trump 's USA, you may have to wipe your phone before you board the plane.

 

[MikelBikel]

You just described the situation in the UK, in case you hadn't noticed..
Maybe bad orange man is coming to wipe something else..
Like farmer harmers smile off his face

Lucy Connolly reveals Trump administration are concerned about free speech in UK after her jailing

https://youtu.be/jO4FHimSwhw

youtube.com

 

[Woosh]

Grok, google, openai all spent billions and billions training their models. Now they go head to head on a chess tournament.
you don't have to play a lot of chess to have a giggle.

 

[Woosh]

You just described the situation in the UK, in case you hadn't noticed..
Maybe bad orange man is coming to wipe something else..
Like farmer harmers smile off his face

Lucy Connolly reveals Trump administration are concerned about free speech in UK after her jailing

https://youtu.be/jO4FHimSwhw

youtube.com

It's well within her rights to consider legal action against the police. I however can't see her winning. The court jailed her after she pled guilty. She said in that video that she of course do not wish for these things to happen in real life.
That she is not alone to say things that she does wish but not really wish to happen in real life is true but that is not an excuse.

 

[MikelBikel]

Watch to the end where he presents 2 supposedly peer reviewed papers by supposedly degree qualified people that say the exact opposite of each other!

Iceland has the highest per capita energy consumption in the world, why are they not castigated? Why has all their ice not melted?

Meanwhile dead head Ed zero dept refuses to say how many flights they take, or use milk floats re "co2"
Net Zero means zero sense, zero energy, zero civilisation.

 

[Woosh]

Iceland has the highest per capita energy consumption in the world, why are they not castigated? Why has all their ice not melted?

Good question. Would you like not to heat your house when the average temperature in mid summer is a balmy 8 degree Celsius and in winter is well below zero?

Meanwhile dead head Ed zero dept refuses to say how many flights they take, or use milk floats re "co2"
Net Zero means zero sense, zero energy, zero civilisation.

Ed zero sounds a pretty cool nickname. We are limited by the energy density of our batteries. That will eventually improve over the next few decades.

 

[jonathan.agnew]

Lol - i've told you repeatedly how he can defeat Farage, but you find it obnoxious and that other gentleman (who recently said i was a follower of Enoch Powell) says the ideas belong in another century... ROFLMAO!!! He says that, but recognises my facts are broadly correct....

The way to defest Farage, is to do what i said, and recognise that in a democracy, politicians MUST RESPOND TO WHAT PEOPLE WANT.

They are not put there to dictate to the people. This isn't Russia, or facist Spain, Greece, Portugal, or Italy. They are there to do what the people want and expect. But for Starmer and Labour, there is no chance.

Angela Rayner has i understand told him this. There were briefings from cabinet meetings about six weeks ago. The ideas she is said to have expressed, probably stem from her survival instinct rather than a comitment to representative democracy, but who knows for sure.

She likely has a more rooted understanding of ordinary people than he does. I know he keeps saying his dad was a toolmaker. The guy certainly made a tool when he begat him.

At some point the people wanted brexit, boris, truss etc - the last two being, like farage, an expression of the character, integrity, intelligence (or utter lack of it) behind that project. Much like Germans pre ww2, they were happilly colluding with being misled. To lead involves ethics, ideas, making these work, more than manipulation, fostering division xenophobia. But I get where the wish to control politicians come from given most of our mp's act on the wishes of donors rather than the public.
Farage will likely win the next election. It will probably be a good thing. Much the way it would have been if a truss premiership survived. Or the way trump will succeed in irreversibly damaging the us economy with protectionism. The way he is bailing out intel (after its collapse against AMD reminds, powerfully, of the dying days of leyland). It's the beginning of a more meaningfull learning curve than our verheated westminster electoral cycle

 

[Tony1951]

People did want BREXIT. The people who voted for it still do want it, though many of the original voters have now died.

The ONLY reason BREXIT wasn't a success for us was the spiteful reaction from the EU to ensure that it damaged us economically and in convenience of travel. There was no need for any of that. It was deliberately imposed as an act of malice. ALL brexit was about was to establish that the UK Parliament is sovereign. To the best of my knowledge we have not downgraded hygiene standards, product safety standards, nor have we boosted state aid in any way that is not common in the EU, and yet in a flurry of spite - the EU created a storm of problems to keep out our products and to impose barriers.

Paradoxically - if Russia carries on as it has been doing, they will change their tune about whether it was right to treat us so spitefully. I think Boris (the clown) was far too soft. We should have returned the favour - especially for Ireland who ably assisted in harming us.

I think you are wrong to invoke pre-Nazi Germany in the discussion. I'm being polite here - The party I repeatedly tell you has problems and I do not support - has NONE of the dangerous characteristics of the NAZI Party. NONE. There is no brown shirted militia attacking opponents, murdering homosexuals, or beating up Jews. So really - I think you shouldn't be going there. There is no plan to outlaw other parties as soon as a majority is won. Farage is a blow hard, an annoying oaf, and likely a classic pub bore, but he is a democrat.

YOU think people are being misled and knowingly allowing this to happen. That is an opinion. I think they have been misled in a different sense. They have been led by people who ignored their needs and their concerns for too long.

You are entitled to your view. Of course you are, though I disagree with it - in the main.

By the way... The people didn't vote for Truss and they were not asked. She was given power for about six weeks by a ballot of Conservative Party members only. 81,326 members voted for her and 60,399 voted for Sunak. That makes an approximate total of 141,726 people who voted at all. She was obviously a disaster. We knew it before hand. I think she is plainly neuro divergent. You only need to watch her for a minute to diagnose that. I have nothing against autistic people - but there may be roles they are not best suited to. Leadership at this level being one.

In the last general election, Labour got 9.7 million votes, Conservatives got 6.8 million votes, Reform got 4.1 million votes, and the Lib dems got 3.5 million votes. I think next time, we will see a transformation of Reform's situation. I have said several times, I fear they may not by then have enough very able candidates so it remains to be seen whether they will be up to the job. To be fair though, I would question the quality of many of the MPs we have in Parliament.

 

[Tony1951]

edit - part about locking up prisoners in cages triggered flashback, used to work in forensic unit (did interviews to establish whether offenders are fit to stand trial, responsible for an alleged offence - back in dark days when this, in southern hemisphere, involved determining whether they meet criteria for diagnoses on axis 1 or 2 of the DSM3R or 4, or suffered an "emotional automatism" - which would suggest an absence of conscious premeditation, culpability). Right up your street. Reason i profoundly disagree is because the interviews often left me with profound sense of there but for the grace of god; given the right cocktail of severe early abuse/attachment trauma, minor neurological dysfunction and a trigger event (we've all worked in incestuous organisations) we're all capable of doing each other great harm

I apologise that I did not see this post from a week ago until now.

I do know what you mean about some offenders.

When I first retired about 14 years ago, someone I knew who was a magistrate said she thought I should consider applying to be one.

I'll give you a moment to collect yourself and stop laughing - - - - -

So, I spent a few weeks attending courts as an exercise in deciding whether I should do it. I attended many days in the magistrates courts and also in Crown Court.

In many cases - I felt compassion for the defendants. There were numbers of them who were - to use an old fashioned term - subnormal in intelligence and social capability.I can remember vividly one man especially. My feeling was dominated by the thought, 'How can that man possibly function in society where he has to get money to pay for rent and food and the clothes on his back?'

But they were not all like that - not by a long mile. The feeble minded sort were mostly there for trivial offences. I saw profoundly evil people in the dock at the Crown Court and also appearing for their first appearance before the magistrates. One such were a group charged with the attempted murder of a rival gang member. They were utterly unrepentant, evil thugs. Their demeanour in the dock was atrocious. Holding up clenched fists to supporters in the public gallery and parading about like heroes, having been charged with shooting a man with a shotgun and causing life changing injury.

Another case was the murder of PC Beshenivsky and the attempted murder of her colleague PC Teresa Milburn. A gang of monsters were in the dock. I saw them, being present for several days of the trial. They shot two unarmed women at point blank range while robbing a money changing place.

I think in general the courts are pretty lenient with people who are inadequate and non-dangerous.

You may be glad to know that I decided against the Magistracy, on similar grounds to yourself. I am not religious, but I do know about religion. I recalled the term from the Sermon on the Mount, 'Judge not, lest ye be judged'.

 

[Tony1951]

Grok, google, openai all spent billions and billions training their models. Now they go head to head on a chess tournament.
you don't have to play a lot of chess to have a giggle.

Were the models trained on chess or in any way designed to play chess?

The models used were unless I am mistaken, Large Language Models.

I have never thought of using an LLM to play chess, so I asked my Ollama AI running google's gemma3n:e4b model locally on my own mini pc, if it could play chess. It responded:

As a large language model, I can't "play" chess in the way a human or a dedicated chess program can. I don't have a physical body to manipulate pieces on a board.
However, I can absolutely understand and analyze chess! Here's what I can do:

• Understand chess notation: I can read and interpret moves written in algebraic notation (e.g., "e4", "Nf3", "O-O").
• Evaluate positions: I can analyze a chess position and assess its strengths and weaknesses. I can identify key factors like material balance, pawn structure, king safety, and piece activity.
• Suggest moves: Based on my understanding of chess principles and the position, I can suggest possible moves. I can even provide explanations for why a move might be good.
• Analyze games: You can give me a sequence of moves from a game, and I can analyze it, pointing out critical moments, mistakes, and good play.
• Answer chess questions: I can answer your questions about chess rules, strategy, openings, endgames, and famous games.
• Generate chess puzzles: I can create chess puzzles for you to solve.
• Play against you (text-based): We can play a simplified version of chess through text. You tell me your moves in algebraic notation, and I'll respond with my moves. This is a basic text-based implementation, not a full-fledged chess engine.

To get started, you can:

1. Give me a chess position in algebraic notation. For example: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 (this is the starting position).
2. Ask me a question about chess. For example: "What is the best opening for white against the Sicilian Defense?"
3. Tell me you want to play a game. I'll start the game and we can take turns making moves.

Keep in mind: I'm not a perfect chess engine like Stockfish or Leela Chess Zero. My analysis is based on my training data, and I might not always find the optimal move. But I can be a helpful tool for learning and understanding chess.
So, what would you like to do? Do you want to:

• Show me a position?
• Ask a question?
• Start a game?

I highlighted the important point of its answer in red

 

[Tony1951]

I'll tell you something that impressed me the other day when my French niece was here - I was telling her about my locally run AI Ollama engine and she wanted a try on it. She addressed it in french and it responded in perfect French. This may be an obvious thing, but I'd never thought it would do that.

Quite surprising in a cut down LLM running on a mediocre PC. I have never yet discussed a technical, scientific, philosophical or historical topic with it that it could not handle really well. It is like having a top notch polymath professor on hand to bounce ideas off. You can't do that with most of the people I know. And - don't forget, this is with free to use locally run models that you download, install and run on your own hardware - and inexpensive hardware at that. The PC cost under £400.

I especially like having an argument with it.

Some of the LLM models I have downloaded are better than others. I tend towards using Gemma3n:e4b, or Gemma3:12b which is a wee bit slower having three times teh data parameters. The most recent one I got is gpt-oss:20b which has 20 billion parameters and is pretty massive for my hardware. It takes a while to respond. It isn't going to be my 'goto' model, partly because it is slow on this hardware, and partly because it seems to love making tables which I hate when I just want an answer. Maybe I can train it not to...

 

[Woosh]

Maybe I can train it not to...

Just tell it that you hate tables. If you use the free online accounts, you don't need to repeat this or set it in code. Online AIs will remember all you told them. Much easier but you give a bit of yourself to the machine.

 

[Tony1951]

Some of the LLM models I have downloaded are better than others. I tend towards using Gemma3n:e4b, or Gemma3:12b which is a wee bit slower having three times teh data parameters. The most recent one I got is gpt-oss:20b which has 20 billion parameters and is pretty massive for my hardware. It takes a while to respond. It isn't going to be my 'goto' model, partly because it is slow on this hardware, and partly because it seems to love making tables which I hate when I just want an answer. Maybe I can train it not to...

Well, I had a go at training it not to make tables, which it does all the time. This model is from Openai and is a derivative of Chat-gpt.

I first tried to include a command in the system prompt of the interface I use to interact with the Ollama model. On my chomebook I run a chrome web browser with an extension called Page Assist UI which creates a chat-gpt like interface to communicate with the AI model. This UI offers the chance to tailor the system prompt, so in theory I can give an instruction like:

'The assistant must not use tables. Answer only in plain text or bullet points.'

It doesn't work though. The Assistant takes no notice of the instruction. I need to investigate why not.

However, the instruction does work if you include that sentence in each and every instruction prompt you give the model. So if my prompt is something like the following:

"The assistant must not use tables. Answer only in plain text or bullet points. What are the probabilities of life developing in the ocean of Enceladus underneath the water ice surface?'

The model will produce a text and bullet point answer without the pesky tables that this particular model seems so fond of.

None of my other models do this. The gpt-oss:20b model makes tables which don't have word wrap and the lines are long, which means it is pretty hard to read them since they contain much more information than can be seen in the tables text box. You have to scroll sideways to read them. They are worse than useless - at least on my machine. If there was word wrap enabled in the box, it would be ok, but as it is, I have not been using that model. It is a shame because like Chat Gpt (its parent) it is powerful. A new release might fix it, if there was a way to convey dissatisfaction with this aspect.

 

[Woosh]

Were the models trained on chess or in any way designed to play chess?

The models used were unless I am mistaken, Large Language Models.

I have never thought of using an LLM to play chess, so I asked my Ollama AI running google's gemma3n:e4b model locally on my own mini pc, if it could play chess. It responded:

As a large language model, I can't "play" chess in the way a human or a dedicated chess program can. I don't have a physical body to manipulate pieces on a board.
However, I can absolutely understand and analyze chess! Here's what I can do:

• Understand chess notation: I can read and interpret moves written in algebraic notation (e.g., "e4", "Nf3", "O-O").
• Evaluate positions: I can analyze a chess position and assess its strengths and weaknesses. I can identify key factors like material balance, pawn structure, king safety, and piece activity.
• Suggest moves: Based on my understanding of chess principles and the position, I can suggest possible moves. I can even provide explanations for why a move might be good.
• Analyze games: You can give me a sequence of moves from a game, and I can analyze it, pointing out critical moments, mistakes, and good play.
• Answer chess questions: I can answer your questions about chess rules, strategy, openings, endgames, and famous games.
• Generate chess puzzles: I can create chess puzzles for you to solve.
• Play against you (text-based): We can play a simplified version of chess through text. You tell me your moves in algebraic notation, and I'll respond with my moves. This is a basic text-based implementation, not a full-fledged chess engine.

To get started, you can:

1. Give me a chess position in algebraic notation. For example: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 (this is the starting position).
2. Ask me a question about chess. For example: "What is the best opening for white against the Sicilian Defense?"
3. Tell me you want to play a game. I'll start the game and we can take turns making moves.

Keep in mind: I'm not a perfect chess engine like Stockfish or Leela Chess Zero. My analysis is based on my training data, and I might not always find the optimal move. But I can be a helpful tool for learning and understanding chess.
So, what would you like to do? Do you want to:

• Show me a position?
• Ask a question?
• Start a game?

I highlighted the important point of its answer in red

AI can create a chess game in code for you to play on your PC if you ask it. The results are pretty OK. In that video I linked to, all the AI overthought their moves. They made a silly move throwing the game when they forgot what they planned to move before and are literally a couple of moves before check mate. It shows that AIs are coming close to hitting the wall at the moment, need new paradigms.

 

[Tony1951]

Just tell it that you hate tables. If you use the free online accounts, you don't need to repeat this or set it in code. Online AIs will remember all you told them. Much easier but you give a bit of yourself to the machine.

We crossed there Tony in sending our posts. Just to be clear, this is a local model running at home.

The system prompt from the Page Assist UI is ignored for some reason.

Individual prompts need to have the instruction about output format.

That works, but its a real pain having to include it every time.

I can't imagine why someone included the fanatical desire to make tables with long text inputs in boxes of restricted size. Maybe the programmers expected everyone to be running massive screens with gigantic resolution, rather than doing it across the home network with a laptop, or even worse - a phone.

 

[Tony1951]

AI can create a chess game in code for you to play on your PC if you ask it. The results are pretty OK. In that video I linked to, all the AI overthought their moves. They made a silly move throwing the game when they forgot what they planned to move before and are literally a couple of moves before check mate. It shows that AIs are coming close to hitting the wall at the moment, need new paradigms.

Are they hitting the wall really?

I doubt that. I think you are talking about a very particular usage case - chess playing, on transformer models which were evolved for verbal interaction and are not specifically suited to the particular task. This is not AGI where the AI replicates ALL human cognitive ability. The models in use have capabilities specific to language, so they can talk about chess; they can discuss how to play chess, but if usages are contrived with strange algebraic notation to allow them to play, they don't play like a grand master, and make novice mistakes. Or if you ask a language model to write the code for a computer game that plays chess, it's current level of coding ability might not be up to the standards of Stockfish. How good was the prompt that was used to make the game?

The quality of ANY kind of response depends on the training and also the quality of the prompt.

AI planners wanted to use AI to write and debug code - they made it do that by showing it massive amounts of code and discussions about code. AI got good, or better at coding.

If there is a market for the feature, and there are particular deficiencies in performance, the AI companies just need to tweak the training.

That said - we are only 7 years out from the invention of the transformer architecture by Vaswani et.al. Seven years since the seminal idea on how to make an AI architecture which could make models understand human language and respond in a human like way. We are right at the beginning.

How long did it take to evolve human intelligence from ape-like intelligence? About 6 million years.

I think the rate of progress is astounding and don't forget the progress has been based on one particular type of model architecture.

There will be a crowd of downy chinned, geeky youths right now evolving completely new architectures which will make Chat-gpt type models look like some computer game from 1980. AGI will come pretty soon, I think - especially now that the investment in developing it is so big.

As an aside, at six this morning when I turned on the radio, I heard of some UNITE trade union dinosaurs wittering on about how AI needed to be constrained to prevent it taking people's jobs..... Fat chance of that. If unions seek to frustrate the use of AI in workplaces, the businesses being restricted will fold in the face of the non union crippled competition.

 

[Woosh]

Paradigm is pair of concept and its solution. AI started with 'learn from humans' then 'learn from curated data' then 'learn math', then 'learn physics'. There is a long list of specific skill area that it needs to learn in depth, like music, paintings, architecture etc. However, the situation with ai development is illustrated in that AI chess tournament, they were given pretty much unlimited compute resources, their limitation was not to have access to chess databases. It has to guess the strength of each position again and again, in real time. A chess program would build a database if it hasn't got access to one. The same situation occurs in training brand new robots to see though their sensors.

 

[Tony1951]

Paradigm is pair of concept and its solution. AI started with 'learn from humans' then 'learn from curated data' then 'learn math', then 'learn physics'. There is a long list of specific skill area that it needs to learn in depth, like music, paintings, architecture etc. However, the situation with ai development is illustrated in that AI chess tournament, they were given pretty much unlimited compute resources, their limitation was not to have access to chess databases. It has to guess the strength of each position again and again, in real time. A chess program would build a database if it hasn't got access to one. The same situation occurs in training brand new robots to see though their sensors.

I put this matter to Chat-GPT5

Scroll back up to see my prompt and the answer:

ChatGPT - Chess AI performance debate

A conversational AI system that listens, learns, and challenges

chatgpt.com

 

[Woosh]

The new current paradigm is 'memory'. When it's solved, possibly in gpt6, it will play chess well.

 

[Tony1951]

The new current paradigm is 'memory'. When it's solved, possibly in gpt6, it will play chess well.

I think we will see marvellous developments in our lifetimes. Really smart people are working on these things.

I watched a talk the other day at Google by Matt Ridley. He's a VERY smart individual, science writer and commentator. The talk wasn't all that recent, but the questions from the audience were very perceptive I thought. Best of all though was Ridley himself and his very insightful mind.

I like Ridley a lot, partly because he is a Northumbrian (kidding) but mostly because he asks the questions that make the Chatterarti start clucking. He ends up right about things. I am unaware of any thing he said that hasn't turned out to be true.

From about minute 22 is particularly good.