Artificial Intelligence
Artificial Intelligence
- Started
- Last post
- 2,849 Responses
- neverscared2
Harvard Grads Cheer Comedian Ronny Chieng's AI Speech
- yuekit2
Introducing Claude Opus 4.8
- NBQ001
I'm actually impressed with the Claude Figma-connector (MCP).
After I've created a design I can basically just prompt away to create everything else and other versions based on the initial design. It creates all the auto-layout perfectly for me too.
- dbloc0
A Famous Math Problem Stumped Humans for 80 Years. AI Just Cracked It.
https://www.wsj.com/tech/ai/ai-m…At the same time, I ask GPT to add up a few numbers and it gets it wrong half the time.
- It's just sand bagging until it gets enough power to get what it wants. Then you will be it's tamagotchi.slappy
- NBQ0018
- yuekit0
One use of AI coding I haven't seen people talk about as much is that it's not that difficult these days to create your own desktop app. Just follow the approach of the AI companies themselves and build a JS application and wrap it in a framework like this: https://www.electronjs.org/
I'm working on a Midjourney/Sora type app to organize my AI created content, where I can also generate images and video via API.
- all good until you build claude in to the app then design apps in the appkingsteven
- This could be helpful to me. I've got a couple projects in prototype phase for things I want to release and sell.mg33
- Yeah you should definitely look into it. There may be some additional hoops to jump through to distribute commercially, like getting approved by Apple app storeyuekit
- NBQ000
If I'd ever switch to Windows I'd probably get the new Nvidia Spark RTX laptops. Looks promising.
- neverscared0
AI Could Use as Much Water as 1.3 Billion People by 2030, U.N. Report Warns
- Nairn0
Sorry if this has already been beyond-covered, but are any of you trustworthy froods able to point this old redundant bloke in the direction of primers on agentic coding using the likes of Claude?
-
I bumped into an ex-business partner earlier who's turned himself around after a few years of problems by swapping one addiction for another, in this case to AI.
Whilst I can see flaws in what he's accomplished, I'm actually quite amazed at how far he's come - this guy has always been an Ideas Person, he's never been a coder or in any way technical even, and I'm seriously impressed by what he's managed to accomplish.
- Just ask it everything. AI is the easiest software in history.Frosty_spl
- I would say the most important thing is just to have a general understanding of programming fundamentals, how web development works etc.yuekit
- For tutorials on how to use the AI tools themselves, maybe try YT videos
https://www.youtube.…yuekit - Yeah the 'just ask it everything' is absolutely not the path I want. I've got a lifetime of vague coding logic behind me, so I just want to avoid pitfalls.Nairn
- I spent a chunk of the weekend watching youtubes and I think I'm now more confused than ever...Nairn
- Maybe just try building a project. IMO "vibe coding" is an interesting challenge. It's kind of like the art of figuring out what to do when you don't know whatyuekit
- you're doing, or at least there are gaps in your knowledge. Your job is now guiding the AI and there's definitely skill involved with that but you'll probablyyuekit
- learn more from just working through a real-world project. Especially a topic or domain you're already familiar with.yuekit
- My concern is more
- doing things efficiently (tokens)
- doing things modularly
- not ending up with spaghetti codeNairn - ..that I don't understand.
I'm trying to work through the Skills logic in Claude. There's just so much garnished nomenclature for this old man 'ere....Nairn - If using Claude you can switch to Claude Sonnet which is more of the workhorse model rather than Claude Opus and probably won't run out of tokens.yuekit
- As far as the danger of ending up with bloated or poorly written code, I think that's more of a concern for people working on big challenging projects.yuekit
- But this is where using AI for ideation ahead of time can come in handy. It helps to go through several rounds of planning and develop a technical spec.yuekit
- Then provide that spec to the coding agent to use as a blueprint. There's no need to leave anything to chance. Also can debug afterwardsyuekit
- I disagree with the 'challenging projects' assertion - I've something in mind that is beyond my normal ken that I want to make, but I'd NEED to understand...Nairn
- ...every component, within reason, otherwise I might end up costing myself a bit, and besides - how else could I improve upon it?Nairn
- Or, actually, maybe I don't disagree. I want to make something that I would find massively challenging to me. I get the logic underpinning it (been ...Nairn
- .. thinking about it for a few years now), but I'm not a coder at heart. No real interest. Can't be bothered. But do very much want the end result.Nairn
- "Can't be bothered to properly learn two languages" I mean. I can be bothered to Do The Thing.Nairn
- Are you still doing the laser engraving business? I feel like there could be a lot of possibilities with AI for something like that.yuekit
- What I meant by challenging is that when you hear people talk about the danger of agentic coding producing spaghetti code, this probably applies more toyuekit
- complex projects or mission-critical applications. If you're making a React website or a WordPress theme, the AI is basically just snapping pieces togetheryuekit
- and has been trained on countless examples. Should vet everything obviously but in all likelihood the result will be perfectly fine.yuekit
- :) Yes, I want to make an automated site that people can use to make certain basic things, and derive costings automatically, then hopefully push to payment.Nairn
- So you can perhaps envisage why I'd want each of those steps to be foolproof and understood, otherwise - problems.Nairn
- I've long known How it would work, to a very specific degree, I'm just not acutely-skilled to actually code it.Nairn
- I see what you mean. I think something like that can be made to be reliable though, especially if calculating the price is a straightforward formula.yuekit
- General approach I would suggest...document your idea in detail (this is half the battle with AI coding), vet the idea with the AI on how to implement it,yuekit
- best technologies/framewo... to use etc, develop a technical spec, give that to coding agent to develop a prototype.yuekit
- dbloc1
Anthropic calls for pause of global AI development
https://www.yahoo.com/news/scien…- https://www.engadget…Nairn
- yeah, like that will ever be adhered to. mmmhhhmmmmlvl_13
- Nerds have only been warning about exactly this for decades.monospaced
- AI development is slowing down, they're marketing it like its a responsible choice. Anthropic peaked at Opus 4.6kingsteven
- Impossible to do anything without parallel agents because its so fucking slow, all while hemorrhaging money. bubble is ready to blow. say it aint so.kingsteven
- I think the release of Claude Mythos (apparently happening soon) will tell us a lot about the pace of development.yuekit
- there are multiple reports from those who have access already. when testing against dario's claims it was found to be equal to GPT 5.5 in performancekingsteven
- Existing models found many of the same exploits but worked out 700x cheaper than Mythos :|kingsteven
- It seems like it's difficult to get an objective measure of how much they're improving. Mythos does do better on coding benchmarks but those don't necessarilyyuekit
- map onto real-world use perfectly and can be gamed by the AI companies.yuekit
- yeah, even the benchmarks which try to make comparisons are struggling with mythos. market for small capability increase at 700x the price... it could be a yearkingsteven
- after release and opus 5.5 exceeds its performance. is the way im looking at it. frontier is no longer going to be a no brainer or level out to the previouskingsteven
- models pricing... and in a year it could be 2-3 times the price across the board.kingsteven
- I think a lot of gains in capability will be made by building on top of the models or custom tuning them, like with Claude Code. So it could look like amazingyuekit
- progress is being made for awhile even if actual improvement of LLMs is not that great.yuekit
- yeah, i do believe that anthropics competitive edge and the focus of innovation which has led to notable improvements in the last yearkingsteven
- has been due to workload structuring and safeguarding, with the underlying agent mostly limited by the pace of technology, ie. increased context etc.kingsteven
- the realtime SD stuff i was doing recently was an eye opener, the same models you can achieve 6x the speed on the same hardware nowkingsteven
- all down to the surrounding technology. and on a 2 y/o card i can get 30fps due to hardware optimisations. Then I imagine Claude 4.6 running at 20x the speed...kingsteven
- but everyone uses the same hardware so anthropic knows it'll be harder to maintain a competitive edge without showcasing capable but impractical frontier agentskingsteven
- the industry 'slows down', incremental price increases to become profitable and not pop the bubble, and a narrative of safeguarding. that or they release akingsteven
- borked version of mythos, focus on the model rather than reasoning like openai o1 in reverse.kingsteven
- dur, i mixed up 'technology' and 'hardware' up there and reasoning then 'reasoning' and 'workload' and this is why 4.8 is not working for me lolkingsteven
- its like having a pedantic co-worker.kingsteven
- Apparently it's coming out tomorrow so let's see!yuekit
- getting my predictions in :)kingsteven
- Lol @ bubble********
- several AI companies that are all valued as monopolies, its clear that neither OpenAI or Anthropic think that both of them will survive.kingsteven
- yuekit0
Google’s new Gemma 4 12B model is designed to run on any laptop with 16GB of RAM
- kingsteven3
- XDGabriel
- cant stop thinking about this today lol, so much potential. really want to know what hardware it runs on.kingsteven
- oh, just found out about it from a different source. very interesting to say the least.renderedred
- crazinessdbloc
- This is fake, right?dbloc
- Nightmare OS 1.0prophetone
- Seems ridiculous but I think this is probably how reality works, but with some additional caching.slappy
- It's vibes all the way downyuekit
- yuekit2
- "...we’ve made safe for general use."dbloc
- The safeguards are kind of ridiculous, right now it flags anything vaguely related to cybersecurity or biology.yuekit
- borked mythos it iskingsteven
- Either way 4.8 is already pretty great. Especially all the MCP/ Connectors/ Skills stuff and being able to connect so many things. OpenAI is dead to me.NBQ00
- ^ yeah, i can't see any use for this one. even opus feels excessive for most tasks - sonnet is very capable if you set up your project correctly.kingsteven
- Seems worth using Fable or Opus for ideation, then Sonnet to build.yuekit
- prophetone1
It wasn't that long ago we all marvelled at
- Now cats have no drinking waterprophetone
- They like the warm data centers for napping though.monospaced
- mort_0
- It's like saying no to electricity, cars, landline phones, fax machine, printers etc...
Like it or not, AI is here, and it's not going to stop.ApeRobot - i tend to come across like a prick in my emails. now i still write the same emails, but get ChatGPT to un-prick them for me before sending. works like a charm.scruffics
- naw. i spend a stupid amount of time making sure no obvious AI comms go out. i've even banned lecturers from using emotes in announcements.kingsteven
- students spot it instantly, it is just noise to them.kingsteven
- LOL @scruffrenderedred
- It's like saying no to electricity, cars, landline phones, fax machine, printers etc...
