Avalon local LLM

[SteamyTea]

I hate coding, even though I do appreciate what can be achieved with it.

But as computer coding is a logical processes, is AI not showing up it's weakness but not being able to write some scripts easily? Or is it that most programming languages are so full of contradictions that the whole industry needs to have a word with itself.

[Pocster]

1 hour ago, SteamyTea said:

I hate coding, even though I do appreciate what can be achieved with it.

But as computer coding is a logical processes, is AI not showing up it's weakness but not being able to write some scripts easily? Or is it that most programming languages are so full of contradictions that the whole industry needs to have a word with itself.

Thats actually an extremely good point. AI coding isn't like mine at all. The style, the way AI usually over engineers. As models are trained off existing code bases then it loses the "human" element of coding i.e. elegance . Ask 50 competent programmers to write pac-man and you'll get 50 packman games - the code though will vary wildly on how they solve issues and do things - infinite variation in fact. The future will be IMHO AI checking other AI written code no human coding. Of course that's what I do now! Chat checks codex suggestions but when chat gets in a mess I literally say STOP!. We then uncommit previous code so we are back to a known state. Though I have not looked still at 1 line of code I can see where it's getting stuck. Then I can direct it better on what to do.

[SimonD]

5 hours ago, SteamyTea said:

so full of contradictions that the whole industry needs to have a word with itself.

That probably sums it up rather nicely. It's random. Having been out of the game for over 2 decades, I'm not massively surprised by the short cuts and poor approaches to design that I'm seeing in programming generally. I've pulled Claude up on that a few times but we have a chat about it and then take an informed decision. Probably one of the best aspects to using AI for coding is the commenting. As long as the prompts specify the extent of commenting, it is miles above what you usually find with human produced code because it takes a lot of effort to think about and formulate comments if you're the programmer. So this makes life so much easier.

Generally though, it's actually one of the good things the AI does and I'm blown away by what it can do, even if it does have its moments. You just have to learn how to compartmentalise the work, slicing it up into smaller components and formulate your prompts.

[SteamyTea]

7 minutes ago, SimonD said:

You just have to learn how to compartmentalise the work, slicing it up into smaller components and formulate your prompts.

Which is what I was taught to do in the early 1980s.

Does it recognise flow diagrams, assuming they are done correctly?

 

  

7 minutes ago, SimonD said:

Probably one of the best aspects to using AI for coding is the commenting. As long as the prompts specify the extent of commenting, it is miles above what you usually find with human produced code because it takes a lot of effort to think about and formulate comments if you're the programmer

Once saw a comment in a database that said "this is shit code"

Edited yesterday at 14:47 by SteamyTea

[Pocster]

33 minutes ago, SteamyTea said:

Once saw a comment in a database that said "this is shit code"

That nothing to what we would do 

 

"turn back , its a (expletive deleted)ing mess"

"this wont work"

"(expletive deleted) knows"

"Jesus this is not going well"

"WTF is this?"

"Why?"

 

etc etc etc. Sega failed us once because they wanted the obj files for a game. It had comemnts in. That means all the bad lanugauge. Took ages to find all the abuse.

The thing was back then when you wrote a game likelyhood of reusing code (unless maybe AI at the time or render engine) was very low. Re invented the wheel each time.

[Pocster]

And just like that it turned into a PS5 demo...

[SimonD]

1 hour ago, SteamyTea said:

Once saw a comment in a database that said "this is shit code"

 

1 hour ago, Pocster said:

That nothing to what we would do 

 

"turn back , its a (expletive deleted)ing mess"

"this wont work"

"(expletive deleted) knows"

"Jesus this is not going well"

"WTF is this?"

"Why?

 

Perhaps I should rephrase to considered commenting that is helpful rather than just crap coming out of the numb brain of a programmer who's had enough and bored!

[Pocster]

24 minutes ago, SimonD said:

 

 

Perhaps I should rephrase to considered commenting that is helpful rather than just crap coming out of the numb brain of a programmer who's had enough and bored!

Once in a certain game if you did a key combo on level 1 my true opinions scrolled up the screen.

 

@SimonD clearly a professional. But when your'e a creative under pressue constantly you aint got time to straighten your tie 

Edited 22 hours ago by Pocster

[SimonD]

7 hours ago, Pocster said:

The future will be IMHO AI checking other AI written code no human coding. Of course that's what I do now!

I was listening to Linus Torvalds speaking about AI during an interview where he was saying how useful AI is and its power to identify bugs and vulnerabilities. But he also said that AI wasn't capable enough to fix them - that needed humans. I kind of agree. The problem is that AI can only look backwards to harvest stuff that already exists and then regurgitate that - it doesn't have any imagination at all and certainly can't see a path or opportunity ahead (I think this is a limitation inherent in the cognitive models used to develop the llms and will probably also seriously limit the function of agi unless they change tack) . To develop this functionality, just imagine the size of the required context window, we'd be building a data centre or 2, if not more, dedicated to each and every user.

[SimonD]

9 hours ago, SteamyTea said:

I hate coding

On this, I should probably mention that I acquired a balance dysfunction following a bad infection. The legacy of this is that I can't spend vast amounts of time in front of a computer screen ever since. So being in front of a screen and trying to read and write code has been virtually impossible for me for more than 2 decades. The advent of AI for doing this has just opened back up a world that had been lost to me. I never really loved coding and always preferred the design and specification of systems but I still used to create stuff. And now I can do that again without horrible symptoms. So from an accessibility perspective it's rather marvelous too.

[SteamyTea]

1 hour ago, SimonD said:

And now I can do that again without horrible symptoms.

Slight confession, I have been playing with an ESP32 tonight. Got it to log some text over my network. Something I have been meaning to do your years.

I did ask ChatGPT to write the Python scripts.  The RPi one was good, almost, the uPython was bollocks, but got it sorted the old fashioned way: traditional web search and 20 fags.

 

 

[Pocster]

1 minute ago, SteamyTea said:

Slight confession, I have been playing with an ESP32 tonight. Got it to log some text over my network. Something I have been meaning to do your years.

I did ask ChatGPT to write the Python scripts.  The RPi one was good, almost, the uPython was bollocks, but got it sorted the old fashioned way: traditional web search and 20 fags.

 

 

Your prompt was the issue . If you don’t know what to ask ….. but that’s how you learn 

Edited 19 hours ago by Pocster

[SimonD]

43 minutes ago, Pocster said:

Your prompt was the issue . If you don’t know what to ask ….. but that’s how you learn 

 

Go on, how did you learn? Did you do the child process which is just chuck in what seemed like a prompt, get a load of garbage and then refine, or did you study prompting first (I can probably guess). Personally I read a few articles copied and pasted some example prompts to see what happened, got bored and just went in to play and found my way that way. I'm sure I've still got a whole load to learn but tbh outputs are generally pretty on point most of the time.

[SimonD]

58 minutes ago, SteamyTea said:

ESP32 tonight.

You've reminded me that I need to set up a Pi for some sensor logging. Carry on trying though - you'll get the outputs you need eventually and its worth the effort. When I first tried, I nearly gave up coz I thought it was all crap and I got rubbish outputs too.

Now I'll get me coat as I'm now starting to hijack @Pocster's thread 😲

[MikeSharp01]

17 hours ago, Pocster said:

As models are trained off existing code bases then it loses the "human" element of coding i.e. elegance

Agree - and it is only trained on code people have shared so there are whole domains, including some languages, it knows very little about in real world terms although it generally has all the theory it is not able to deploy it, in its word, 'cleanly'! 

[SteamyTea]

7 hours ago, SimonD said:

Carry on trying

Well it logged all night and seemed to do what I wanted it to do.

 

[MikeSharp01]

7 hours ago, SimonD said:

Did you do the child process which is just chuck in what seemed like a prompt, get a load of garbage and then refine, or did you study prompting first (I can probably guess).

Creativity, for people who are no longer children, is all about rediscovering your inner child and dumping all your baggage so it does not constrain your thinking.

[Pocster]

9 hours ago, SimonD said:

Go on, how did you learn? Did you do the child process which is just chuck in what seemed like a prompt

You mean learn what prompt to give llm what I want ? . Firstly I usually know what I want. “Python code using three.js fragment shader to render specular highlights on mesh “ . It’s still quite vague but specified enough to be a good start . The secret as you pointed out is small steps this way you’ve more chance of incremental progress . My harness is tight via my menu system . So ; no more than editing 4 files , no seds per file greater than 100 lines etc . Output file dumps all errors . The best improvement was prompt must start with a router . E.g AVALON_ROUTER: local scout only , codex cloud scout suggest patch , patch only etc . Chat forgets so it doesn’t always put this in the prompt . So the menu stops if any of these are wrong or missing . Sticks why it stopped into output file which of course goes back to chat . This tends to keep chat on track most the time .

[Pocster]

1 hour ago, MikeSharp01 said:

Creativity, for people who are no longer children, is all about rediscovering your inner child and dumping all your baggage so it does not constrain your thinking.

This is what llm allows in coding . Limitation isn’t your coding or skills anymore . Knowledge of what you want and how it can be achieved is sufficient. Amazing time to be alive .

[SteamyTea]

Well I just rebooted the ESP and it has all stopped working.

Job for a rainy day.

[SimonD]

3 hours ago, MikeSharp01 said:

Agree - and it is only trained on code people have shared so there are whole domains, including some languages, it knows very little about in real world terms although it generally has all the theory it is not able to deploy it, in its word, 'cleanly'! 

 

1 hour ago, Pocster said:

You mean learn what prompt to give llm what I want ? . Firstly I usually know what I want. “Python code using three.js fragment shader to render specular highlights on mesh “ . It’s still quite vague but specified enough to be a good start . The secret as you pointed out is small steps this way you’ve more chance of incremental progress .

 

1 hour ago, Pocster said:

Chat forgets so it doesn’t always put this in the prompt .

 

So the absolute bonkers thing here is that last night I had a chat with Claude about my outdated and pretty rusty approach to development. And I got some feedback. Here are some examples:

First Claude butters me up:

"Your domain knowledge is exceptional and it shows. The handover documents are some of the best I see....."

To let me down gently:

"You build features faster than you solidify foundations....."

"Bugs are fixed reactively rather than by closing the class."

"Session continuity relies heavily on handover documents rather than code structure."

 

"The "upload current files first" pattern is correct — but you sometimes skip it."

 

"You could use me earlier in the design phase. Most sessions seem to start with "here's a feature, let's build it." You'd get more value from 10 minutes upfront: "here's what I'm thinking, what are the architectural consequences?"

"The emergent style has worked well to get here. The shift you probably need now is from "make it work" to........which means investing a bit more in the boring stuff: "

Now I'm building out a modified set of prompts combined with project references to hand the responsibility of dealing with these things to Claude, with Claude's help!

 

2 hours ago, Pocster said:

This is what llm allows in coding . Limitation isn’t your coding or skills anymore . Knowledge of what you want and how it can be achieved is sufficient. Amazing time to be alive .

It is genuinely amazing. What I've also done is gone through with Claude the best pattern to use with handover between the reasoning and coding tools - so it takes it beyond just the coding. It totally blows me away how powerful this can be. Just a shame it can't help me physically in my day job!

[Pocster]

21 minutes ago, SimonD said:

 

 

 

So the absolute bonkers thing here is that last night I had a chat with Claude about my outdated and pretty rusty approach to development. And I got some feedback. Here are some examples:

First Claude butters me up:

"Your domain knowledge is exceptional and it shows. The handover documents are some of the best I see....."

To let me down gently:

"You build features faster than you solidify foundations....."

"Bugs are fixed reactively rather than by closing the class."

"Session continuity relies heavily on handover documents rather than code structure."

 

"The "upload current files first" pattern is correct — but you sometimes skip it."

 

"You could use me earlier in the design phase. Most sessions seem to start with "here's a feature, let's build it." You'd get more value from 10 minutes upfront: "here's what I'm thinking, what are the architectural consequences?"

"The emergent style has worked well to get here. The shift you probably need now is from "make it work" to........which means investing a bit more in the boring stuff: "

Now I'm building out a modified set of prompts combined with project references to hand the responsibility of dealing with these things to Claude, with Claude's help!

 

It is genuinely amazing. What I've also done is gone through with Claude the best pattern to use with handover between the reasoning and coding tools - so it takes it beyond just the coding. It totally blows me away how powerful this can be. Just a shame it can't help me physically in my day job!

Exactly how I use chat gpt. These llm's are programmed to be passively nice to us dumb humans so equally I get completements on how I treat it like a softeare engineeer and not a toy. I upload code to it (that it wrote) it then blames me as though I wrote it and then it improves or suggests alterations to my (its) code. It admit it doesnt follow hard grounded rules even when commited to memory hence the ridged menu/harness

[Pocster]

I also notice chat seems to offer default simple solutions. So if I ask for a 'face' as I've done. I did give some detail but I end up with a crude basic version. Even if I ask for an authentic water effect and specify it I still get a very basic version. I queried why after I had spec'd it and it said because most people don't want the full thing. I asked for examples. People asking for "a minecraft game" - is obviously a massive project so it gives a very crude simplistic response - because apparently people wont specify exact detail for more advancement. I assume this is where the "toy" reference comes from. Chat seems to believe 99% of coding tasks are simplistic and non challenging "fix this bug" , "refactor this" . Of course you could argue that is not a chatbot's job to be a claude complete "coding solution" - but as I said for 20 quid a month effectively unlimited its the best value for money ever.

[SimonD]

33 minutes ago, Pocster said:

I upload code to it (that it wrote) it then blames me as though I wrote it and then it improves or suggests alterations to my (its) code. It admit it doesnt follow hard grounded rules even when commited to memory hence the ridged menu/harness

Yes, that's exactly it. It drives me a bit mental, especially when it then compliments itself on the code it's created itself! And when I find a bug and it's going round in circles there's been a few time when I said we've had this problem before, haven't we! It then goes oh yes, found it! and then just sorts the instance, rather than the class, like it's telling me. And also the handover docs are absolutely essential otherwise it goes even further off piste!

 

 

42 minutes ago, Pocster said:

I also notice chat seems to offer default simple solutions. So if I ask for a 'face' as I've done. I did give some detail but I end up with a crude basic version. Even if I ask for an authentic water effect and specify it I still get a very basic version. I queried why after I had spec'd it and it said because most people don't want the full thing. I asked for examples. People asking for "a minecraft game" - is obviously a massive project so it gives a very crude simplistic response - because apparently people wont specify exact detail for more advancement. I assume this is where the "toy" reference comes from. Chat seems to believe 99% of coding tasks are simplistic and non challenging "fix this bug" , "refactor this" . Of course you could argue that is not a chatbot's job to be a claude complete "coding solution" - but as I said for 20 quid a month effectively unlimited its the best value for money ever.

Fair comment. I certainly used to get the simple easy response that's only half there until I set up my full project with memory, files, and instructions. And as you say most people won't get past the initial crude request. That's not helped by the abundance of YouTube content telling people how they can build a multi-billion dollar no-code app in 5 minutes. People will try, then find there's a steep learning curve, especially if they've never built software before, and give up after another 5 minutes and instead try the next no-code promise for $10k/month passive income and on to the next one. 

I was quite shocked when I realised that I'd been working on my project since last August and it's June already and I've gone through some right challenges in that time. Now, progress is amazing and I haven't needed half a million quid and a team of software developers in Bangalore (or maybe these are the real homunculus hiding in chat?), so that's incredible for as you say 20 quid a month! And by all accounts they're promising reduced token prices now too!

[Pocster]

10 minutes ago, SimonD said:

Yes, that's exactly it. It drives me a bit mental, especially when it then compliments itself on the code it's created itself! And when I find a bug and it's going round in circles there's been a few time when I said we've had this problem before, haven't we! It then goes oh yes, found it! and then just sorts the instance, rather than the class, like it's telling me. And also the handover docs are absolutely essential otherwise it goes even further off piste!

 

 

Fair comment. I certainly used to get the simple easy response that's only half there until I set up my full project with memory, files, and instructions. And as you say most people won't get past the initial crude request. That's not helped by the abundance of YouTube content telling people how they can build a multi-billion dollar no-code app in 5 minutes. People will try, then find there's a steep learning curve, especially if they've never built software before, and give up after another 5 minutes and instead try the next no-code promise for $10k/month passive income and on to the next one. 

I was quite shocked when I realised that I'd been working on my project since last August and it's June already and I've gone through some right challenges in that time. Now, progress is amazing and I haven't needed half a million quid and a team of software developers in Bangalore (or maybe these are the real homunculus hiding in chat?), so that's incredible for as you say 20 quid a month! And by all accounts they're promising reduced token prices now too!

(expletive deleted)s me off when I see youtube chatgpv vs opus vs fable all given the same prompot to "write a flappy birds 3d game" and thats a test!. Amazing any of them produce anything at all but (expletive deleted) me - it's not a TOY!. It's super powerful.