Artificial Intelligence (AI)

[Trav Le Bleu]

7 minutes ago, leicsmac said:

But in all seriousness, fair point. I think it's wise to know though that the higher the power of a new invention, the greater degree of both good and bad it can cause.

Like the Fleshlight?

Edited 11 August 2025 by Trav Le Bleu

[leicsmac]

On 08/08/2025 at 17:46, leicsmac said:

I think right now there really needs to be more "let's find cancers that were impossible to find otherwise" and less "what if we make the Dead Internet Theory real and shred the fabric of society through half truths while being as avaricious about it as possible, techbros?" about it all, tbh. 

 

There are horrific poisoned excuses for recipes out there that people are cooking because they're being convinced that AI output is reliable. 

Score one for the former category!

 

https://www.bbc.co.uk/news/articles/cgr94xxye2lo

 

Artificial intelligence has invented two new potential antibiotics that could kill drug-resistant gonorrhoea and MRSA, researchers have revealed.

The drugs were designed atom-by-atom by the AI and killed the superbugs in laboratory and animal tests.

The two compounds still need years of refinement and clinical trials before they could be prescribed.

But the Massachusetts Institute of Technology (MIT) team behind it say AI could start a "second golden age" in antibiotic discovery.

[urban.spaceman]

Elon's been made to look like a right tit by his own AI, which I'm sure he'll change to deliver alternative facts very soon. 

 

[Grebfromgrebland]

I have to say that I find grok to be the most trustworthy AI despite all the noise about it. I often ask different AI the same thing and Grok is the least baised.

 

Especially on current affairs.

[urban.spaceman]

google’s Veo video generator is free for the weekend it seems, though with a daily quota limit of about 5 or so videos. 

 

It’s insane. We’re ****ed. 

[The Blur]

A bit of a light relief..

 

https://www.instagram.com/reel/DNr6orn2ASd/?igsh=OTY4eG05NDhzMXdu

 

Had a right good chuckle while watching the video.  

[Trav Le Bleu]

 

Thought provoking.

 

Especially when you realise that almost all our technological advances have been somehow militarised.

[Sampson]

Good video. Scares the crap out of me tbh. 80% of the data is good but it just makes up 20% of the data and then humans use it in videos, websites, articles and even research papers, so it becomes “truth”. Then when AI next pulls data it pulls from this data and uses the new human made things as the source. Creating a vicious cycle of fake “truth”.

 

4 years of this shit before the 2029 elections will be frightening.

 

 

Edited 13 October 2025 by Sampson

[brookfox]

24 minutes ago, Sampson said:

Good video. Scares the crap out of me tbh. 80% of the data is good but it just makes up 20% of the data and then humans use it in videos, websites, articles and even research papers, so it becomes “truth”. Then when AI next pulls data it pulls from this data and uses the new human made things as the source. Creating a vicious cycle of fake “truth”.

 

4 years of this shit before the 2029 elections will be frightening.

 

 

I dunno, kind of looking forward to the moral outcry and calls for Farage to resign when he’s caught having an affair with Tilly Norwood.

[Zear0]

Gemini 3.0 and it's integration with Colab is very very good for those who've not tried it yet.

[urban.spaceman]

Notebook LM is impressive. 

[Zear0]

2 hours ago, urban.spaceman said:

Notebook LM is impressive. 

It's class.  Was doing an MSc. last year and chucked all the notes in there and was great for study.  More recently I've dumped every single Aldi recipe in there to choose my weekly shop and collate the ingredients for shopping list (AI unlocking true laziness)

[urban.spaceman]

15 hours ago, Zear0 said:

It's class.  Was doing an MSc. last year and chucked all the notes in there and was great for study.  More recently I've dumped every single Aldi recipe in there to choose my weekly shop and collate the ingredients for shopping list (AI unlocking true laziness)

I just put 20 pages of an unfinished screenplay in it and within minutes it gave me a video overview of the plot but also an audio ****ing podcast style discussion of the script with three different AI voices BUT THEN YOU CAN INTERRUPT AND ASK THEM A QUESTION ABOUT IT AND THEY ANSWER BACK. Absolutely insane. 

[RoboFox]

As a music producer who's beginning to garner attention, it's made infinitely more difficult by the sheer amount of AI-generated slop that gets uploaded to DSPs every day. We've now seen fully-AI produced "music" top Spotify charts in a few different genres and they're killing it in streaming numbers. 

 

And I've seen some of these thick ***** try to argue that "writing a prompt is a skill" when I and many others have dedicated our spare time over many, many years to getting good at what we do. 

 

I had opportunities to do this years ago but my procrastination and insecurity held me back and I really wish that wasn't the case. 

 

Gemini is GREAT as a tool for research when I'm trying to market my music, it's also fantastic at helping tackle troubleshooting and technical aspects of music and content production software but that's really it. Social media is very much part of the game now and every day I see fellow artists in my scene posting low-effort AI content. And it has a knock on effect to everything: Mastering engineers? Can be done with AI. Cover artwork? AI. Marketing and content generation? AI. It's boring. 

 

Look across Reddit and musician spaces across social media and everyone will tell you using AI for music, artwork or anything else is an immediate turn off, but the truth is for 99% of punters they really don't care. 

 

Some DSPs are half-heartedly forcing creators to declare if something is AI-assisted or not, and some are even outright banning it - but the horse has bolted. And in truth, why would giants like Spotify care? A stream is a stream and the less money they have to pay out, the more profit they make. 

 

Edited 27 November 2025 by RoboFox

[leicsmac]

Saw this earlier. An accurate summation of the current state?

 

 

[danny.]

1 hour ago, leicsmac said:

Saw this earlier. An accurate summation of the current state?

 

 

Yup. And then ask yourself what are the tens of millions of people in this country going to do once we’re at the final stage in that diagram? And why are we importing millions more and urging to have millions more babies. To do what? We won’t have an economy. 

[Grebfromgrebland]

1 hour ago, danny. said:

Yup. And then ask yourself what are the tens of millions of people in this country going to do once we’re at the final stage in that diagram? And why are we importing millions more and urging to have millions more babies. To do what? We won’t have an economy. 

It's not always about immigration FFS. I'm beginning to think you're a bot.

[leicsmac]

2 hours ago, danny. said:

Yup. And then ask yourself what are the tens of millions of people in this country going to do once we’re at the final stage in that diagram? And why are we importing millions more and urging to have millions more babies. To do what? We won’t have an economy. 

Well, that's pretty simple to me. 

 

Either society reorganises drastically in the way it is required to, or society collapses.

 

And that will be a global thing, migration is an adjacent matter here. 

[danny.]

1 minute ago, leicsmac said:

Well, that's pretty simple to me. 

 

Either society reorganises drastically in the way it is required to, or society collapses.

 

And that will be a global thing, migration is an adjacent matter here. 

It's not a global thing as not every country will use AI to the same level

[leicsmac]

1 minute ago, danny. said:

It's not a global thing as not every country will use AI to the same level

Patently so, but the economic and upheaval of so many "first world" countries hitting that bar at roughly the same time will be felt everywhere. 

[Sampson]

3 hours ago, danny. said:

Yup. And then ask yourself what are the tens of millions of people in this country going to do once we’re at the final stage in that diagram? And why are we importing millions more and urging to have millions more babies. To do what? We won’t have an economy. 

On the flip side, if AI does create a world with no jobs as AI can do everything and where everyone is essentially just on a welfare state being cared for by AI, what’s the point even of money and nations at that point? If we don’t need to pay anyone because AI does everything for us. 

[leicsmac]

11 minutes ago, Sampson said:

On the flip side, if AI does create a world with no jobs as AI can do everything and where everyone is essentially just on a welfare state being cared for by AI, what’s the point even of money and nations at that point? If we don’t need to pay anyone because AI does everything for us. 

And there lies the big struggle: between that mindset and that of tbf self-interested, tribal one who sees it simply as a tool to further the power of him and his. 

 

That struggle has always been there, with every new advancement. But this one has the potential to be a true world-changer, either way. 

[Chelmofox]

10 hours ago, leicsmac said:

Saw this earlier. An accurate summation of the current state?

 

 

That's assuming that the current pathway gets us to AGI. 

 

I'm yet to be convinced. Even the best models fail at the simplest tasks (try and get one to translate content at scale but demand it cant go over certain character limits  - its hell). 

 

The newest Generative models get more interesting for sure.  For all the sycophantic posting what you see for Gemini 3, GPT5+ etc, they are largely doing the same things slightly better (geting sick of seeing a 'Ive just created a Netflix clone' post!). I was vibe coding a throw away project today using 2 LLMs, and both only got key things right on the 3rd/4th attempt (i actually gave up with Gemini). It wasnt a simple scenario, but each time they both initially called deprecated methods in libraries that were owned by the LLM's vendors. Anything that has level of system integration or interdependencies is pretty hard for GenAI. I cant imagine just giving that task to anyone for a long time.

 

I am working closely with a Google team building out an Agentic solution for our business.  Its been a fascinating project and we are working with alpha versions of agentic toolkits. Some of what we can do absolutely shocks me. Some of what the agents can appear to understand has left me gobsmacked. We've produced things in minutes that has taken our teams days to put together, and we are only just starting. But, you cannot stop these things going into what we call 'Darren mode', or Hallucinating and being absolutely dumb. I've sat with the Google team whilst we witnessed an Agent (using the latest and greatest) just make any old data up and pretend to run code, even though its core prompts forbid it and the framework caught it. When we interacted with it is completely bullsh**ted us. We then spun up a new instance of the Agent and it ran perfectly. Guys from Google just shrugged their shoulders. We have to build non GenAI guardrails around this, which is fine, but is to some degree defeating the purpose. Im not sure i want GenAI agents managing my banking transactions.

 

Even using the very latest (best ever!) model Google has made available, I still cant see this all running without people running the shop, and this just being something that helps things run faster.  People will have to pick up the slack for a long time in my opinion. 

 

The next versions of the same models are likely going to be bigger / bolder / costlier versions of the same thing. And i think that is going to carry on until either something goes pop (maybe OpenAI is the first to drop?) meaning efficiency becomes the key driver, or someone has something new which renders GenAI obsolete and the pathway to AGI redirects (im not sure there is any evidence of this). The problem then is all the investment in GenAI has likely been a gigantic white elephant, and many will be looking to get their money back.

 

I suppose that's a long way of me saying 'i don't know'. What i can say, for all the impressive outputs we've been able to do with this project, it's been the least satisfying thing ive ever worked on, and certainly stressed me out the most. I don't think im alone in thinking that. Playing around with it is fun, trying to get it to be a reliable asset for your business - tough.

Edited 27 November 2025 by Chelmofox

[Sampson]

14 hours ago, Chelmofox said:

That's assuming that the current pathway gets us to AGI. 

 

I'm yet to be convinced. Even the best models fail at the simplest tasks (try and get one to translate content at scale but demand it cant go over certain character limits  - its hell). 

 

The newest Generative models get more interesting for sure.  For all the sycophantic posting what you see for Gemini 3, GPT5+ etc, they are largely doing the same things slightly better (geting sick of seeing a 'Ive just created a Netflix clone' post!). I was vibe coding a throw away project today using 2 LLMs, and both only got key things right on the 3rd/4th attempt (i actually gave up with Gemini). It wasnt a simple scenario, but each time they both initially called deprecated methods in libraries that were owned by the LLM's vendors. Anything that has level of system integration or interdependencies is pretty hard for GenAI. I cant imagine just giving that task to anyone for a long time.

 

I am working closely with a Google team building out an Agentic solution for our business.  Its been a fascinating project and we are working with alpha versions of agentic toolkits. Some of what we can do absolutely shocks me. Some of what the agents can appear to understand has left me gobsmacked. We've produced things in minutes that has taken our teams days to put together, and we are only just starting. But, you cannot stop these things going into what we call 'Darren mode', or Hallucinating and being absolutely dumb. I've sat with the Google team whilst we witnessed an Agent (using the latest and greatest) just make any old data up and pretend to run code, even though its core prompts forbid it and the framework caught it. When we interacted with it is completely bullsh**ted us. We then spun up a new instance of the Agent and it ran perfectly. Guys from Google just shrugged their shoulders. We have to build non GenAI guardrails around this, which is fine, but is to some degree defeating the purpose. Im not sure i want GenAI agents managing my banking transactions.

 

Even using the very latest (best ever!) model Google has made available, I still cant see this all running without people running the shop, and this just being something that helps things run faster.  People will have to pick up the slack for a long time in my opinion. 

 

The next versions of the same models are likely going to be bigger / bolder / costlier versions of the same thing. And i think that is going to carry on until either something goes pop (maybe OpenAI is the first to drop?) meaning efficiency becomes the key driver, or someone has something new which renders GenAI obsolete and the pathway to AGI redirects (im not sure there is any evidence of this). The problem then is all the investment in GenAI has likely been a gigantic white elephant, and many will be looking to get their money back.

 

I suppose that's a long way of me saying 'i don't know'. What i can say, for all the impressive outputs we've been able to do with this project, it's been the least satisfying thing ive ever worked on, and certainly stressed me out the most. I don't think im alone in thinking that. Playing around with it is fun, trying to get it to be a reliable asset for your business - tough.

That’s a really interesting insight to me thank you.

 

It’s the “Darren mode” as you say and the AI patently lying against its own programming and Google guys just shrugging the shoulders bit that scares me though. Nevermind about your bank would use it, I’m frightened governments, judges and lawyers will use it and decide the fates of people’s lives and the direction society goes in based on AI that openly makes stuff up and lies despite being programmed not to and we don’t understand why. 

Edited 28 November 2025 by Sampson

[danny.]

The amount of gaslighting and lies from LLM is wild now, I’m sure it’s got worse in the last 9-12 months too.