# AI mega thread **URL:** \ **Category:** General\ **Tags:** ai\ **Created:** [October 7, 2026, 11:48pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743 "2026-10-07T23:48:33Z")\ **Posts on this page:** 20\ **Page:** 7
**Author:** ![rexbeng](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/rexbeng/32/356_2.png) [@rexbeng](https://forum.cpcwiki.eu/u/rexbeng)\ **Post date:** [October 11, 2026, 5:54pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/122 "2026-10-11T17:54:10Z")
I disagree. It is an interesting topic and there is plenty more to say, on either side of the argument. ---
**Author:** ![reno08](https://forum.cpcwiki.eu/letter_avatar_proxy/v4/letter/r/48db29/32.png) [@reno08](https://forum.cpcwiki.eu/u/reno08)\ **Post date:** [October 11, 2026, 6:06pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/123 "2026-10-11T18:06:12Z")
What do you not understand about “the guy is using AI without disclosing it, to argue that all use of AI should be disclosed” ? Anyway, can you see how fantastically this decision of pandering to anti-AI hysteria is already working out for camaraderie in the forum ? Told you so early in the thread. Have fun moderating the highly productive “how much AI is too much, is this AI or not, hey he didn’t disclose” endless bickering. Within a couple of years either you’ll be sick of it and reverse the decision, or things will have calmed down because everyone just stopped using the tags. And many people will be ashamed of their posts. Guaranteed. In the meantime, enjoy being on the wrong side of history, I’ll be watching from the sidelines and hope you come to your senses because this was a great CPC resource 👋 ---
**Author:** ![Gryzor](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/gryzor/32/7_2.png) [@Gryzor](https://forum.cpcwiki.eu/u/Gryzor)\ **Post date:** [October 11, 2026, 6:18pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/124 "2026-10-11T18:18:23Z")
> [@reno08](#): > > What do you not understand about “the guy is using AI without disclosing it, to argue that all use of AI should be disclosed” ? Did he say that about all use everywhere? I think it’s pretty evident we’re taking about developing. > [@reno08](#): > > Within a couple of years either you’ll be sick of it and reverse the decision, or things will have calmed down because everyone just stopped using the tags So what? > [@reno08](#): > > people will be ashamed of their posts. Guaranteed. Lol. You seem to think that we discovered AI yesterday. Meanwhile, forgetting CPC ELIZA, I happened to find (only today, interestingly) an early python experiment of mine from 2019. ---
**Author:** ![rexbeng](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/rexbeng/32/356_2.png) [@rexbeng](https://forum.cpcwiki.eu/u/rexbeng)\ **Post date:** [October 11, 2026, 6:34pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/125 "2026-10-11T18:34:28Z")
That was hilarious! Have a great two-years, see you in Fall 2028! 🥳 ---
**Author:** ![Gryzor](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/gryzor/32/7_2.png) [@Gryzor](https://forum.cpcwiki.eu/u/Gryzor)\ **Post date:** [October 11, 2026, 6:42pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/126 "2026-10-11T18:42:00Z")
So I asked Gemini (3.8 Flash, high effort)to summarize his views. For giggles, you understand. The reply: > reno08 suffers from a textbook case of **lazy historical analogy** combined with **purism-baiting**. > > In short, his posts boil down to two classic debate sins: > > ### 1. The 1859 Photography Reflex > > When Rhino points out that generative tools risk devaluing genuine human craft, reno08 immediately reaches for the standard tech-evangelist playbook: dusting off Charles Baudelaire’s 1859 Salon review to claim painters said the exact same thing about photography. > > It is the internet’s favorite canned response to any critique of automation: _“You’re just a 19th-century Luddite yelling at cameras.”_ As rexbeng and Rhino immediately point out, comparing a mechanical recording device (which spawned entirely new artistic movements and never replaced painting) to an automated synthesis machine misses the mark by several miles. > > ### 2. The Slippery-Slope Fallacy > > Across the discussion, his other favorite trick is the _reductio ad absurdum_: _“Why stop at AI? Why not ban everything not coded in pure assembly on real hardware with OCP Art Studio?”_ > > He deliberately blurs the line between modern developer tools (cross-compilers, trackers, IDEs) and generative systems that generate the assets or logic for you. > > ### The Real Blind Spot > > He forgets why people spend their evenings messing around with 40-year-old Amstrads in the first place. Nobody is on CPCWiki for maximum corporate output or industrial efficiency. The entire point of the retro scene is the human effort, the limitation, and the craft. Treating an 8-bit hobby forum like a Silicon Valley disruption seminar completely misses the room. ---
**Author:** ![reno08](https://forum.cpcwiki.eu/letter_avatar_proxy/v4/letter/r/48db29/32.png) [@reno08](https://forum.cpcwiki.eu/u/reno08)\ **Post date:** [October 11, 2026, 6:54pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/127 "2026-10-11T18:54:31Z")
Well, while some people are solving century-old maths problems or simply just porting half of the N64 library to modern Windows with AI, you still haven’t managed to migrate your old forum to the new one so yeah, pretty sure you’ve got lots to discover still. Also, show us your Gemini prompt champ ! ---
**Author:** ![cpcitor](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/cpcitor/32/321_2.png) [@cpcitor](https://forum.cpcwiki.eu/u/cpcitor)\ **Post date:** [October 11, 2026, 6:56pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/128 "2026-10-11T18:56:25Z")
Gentlemen please calm down. 😌 History is happening now. The new forum features (including the automatic translation peppered with em-dashes that post author never asked for) surprise people. Irony may have been used in recent posts. May I remind you that irony carries the risk of being taken literally, leaving the author completely misunderstood, or even taken for a boor? ⚡ I would recommend always re-reading one’s text before posting, considering that statistically, some other persons _will_ (not might, will) read it quickly and miss what you meant. Consider what happens if taken literally. Perhaps include signals like “Warning, upcoming use of friendly irony. 😉”. Or simplify. (I try and not always succeed. 😅 ) ☮ And yes, what we write now will be dated quickly. ---
**Author:** ![Gryzor](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/gryzor/32/7_2.png) [@Gryzor](https://forum.cpcwiki.eu/u/Gryzor)\ **Post date:** [October 11, 2026, 6:58pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/129 "2026-10-11T18:58:29Z")
> [@reno08](#): > > still haven’t managed to migrate your old forum Oh my, nothing like an insult to give gravitas to your argument​😂 > [@reno08](#): > > you’ve got lots to discover still. No shit, Sherlock? ---
**Author:** ![rexbeng](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/rexbeng/32/356_2.png) [@rexbeng](https://forum.cpcwiki.eu/u/rexbeng)\ **Post date:** [October 11, 2026, 7:08pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/131 "2026-10-11T19:08:08Z")
Meanwhile [https://terrytao.wordpress.com/2026/10/06/changes-to-the-erdos-problems-web-site/](https://terrytao.wordpress.com/2026/10/06/changes-to-the-erdos-problems-web-site/) ---
**Author:** ![Gryzor](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/gryzor/32/7_2.png) [@Gryzor](https://forum.cpcwiki.eu/u/Gryzor)\ **Post date:** [October 11, 2026, 7:09pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/132 "2026-10-11T19:09:11Z")
> [@rexbeng](#): > > Meanwhile Very interesting read. What a Luddite. ---
**Author:** ![Anthony\_Flack](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/anthony_flack/32/35_2.png) [@Anthony\_Flack](https://forum.cpcwiki.eu/u/Anthony_Flack)\ **Post date:** [October 11, 2026, 9:46pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/133 "2026-10-11T21:46:17Z")
I think the automatic translation on the forum is neat, although I guess it has the potential to open up hitherto-untapped streams of inter-scene hostility as everyone can argue in their native language. But it’s cool, it works just like Douglas Adams’ Babel fish. Auto-translation of Youtube videos is great too. And I did wonder why not use an AI to migrate all the old forum posts to the new forum. The effects on the scene remain to be seen. Obviously a lot of possibilities have opened up. But being able to do something without effort radically alters the value of the thing. People who previously lacked the patience and attention to make software can now create games without patience or attention. Other people who put the effort in will still be able to make worthwhile things, but we are going to be drowning in careless slop, I don’t doubt it. Because it can be turned out so quickly. And that will affect the value of everything. Does it reduce my desire to code by hand…? Maybe? And things I have done, I figure looking back that people may just assume it was all made by AI anyway. ---
**Author:** ![reno08](https://forum.cpcwiki.eu/letter_avatar_proxy/v4/letter/r/48db29/32.png) [@reno08](https://forum.cpcwiki.eu/u/reno08)\ **Post date:** [October 11, 2026, 10:10pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/134 "2026-10-11T22:10:54Z")
Ok man, @cpcitor is right, we should calm down. I am sincerely sorry for the personal jabs. I feel strongly about this and am sleep deprived, which isn’t a great combination for sensible posting. However, for giggles… never again bring a knife (Gemini 3.8 Flash) to a gun fight. Ladies & gentlemen, here is Opus 5.5 ! 😆 > **Gemini 3.8 Flash suffers from a textbook case of lazy pattern-matching combined with not reading the thread.** > > #### 0. The admin’s form fails the admin’s own post > > Gryzor built a mandatory AI-disclosure form for this forum. One of its fields asks you to list the models used, the specific tools, and **where human intervention was required**. His post #126 lists the model and nothing else: no prompt, no account of human intervention. He was asked for the prompt directly in #127 and never posted it. > > The instruction was supposedly “summarize his views”. The output contains no summary at all, only a prosecution. Either the prompt wasn’t “summarize”, or Gemini can’t follow one. > > And in 603 #21, the same admin said he never passes off an LLM’s work as a final product. #126 is raw LLM output, posted unedited, as a contribution to a debate. > > #### 1. The “1859 Photography Reflex”: Gemini concedes the case and doesn’t notice > > - **It gives away the conclusion.** Gemini says photography “spawned entirely new artistic movements and never replaced painting”. That **is** reno08’s thesis: the fear was overblown and new forms emerged. You can’t cite the happy ending of an analogy as proof the analogy is wrong. > - **“A single letter by a single person” is false.** Rhino said this in #61, and Gemini accepted it. It was a published Salon review by the era’s most influential critic. It also wasn’t isolated. The 1862 petition was signed by Ingres, Flandrin, Fleury-Richard, Nanteuil, Troyon, Bida, Isabey and Puvis de Chavannes (Léon Cogniet and Delacroix refused) to block any assimilation of photography to art; the question arose in a copyright case between the photographers Mayer and Pierson. In 1863 Flandrin wrote that he greatly feared photography had dealt a death blow to Art. [larousse](https://larousse.fr/encyclopedie/peinture/photographie_et_peinture/153827) [sahapedia](https://www.sahapedia.org/index.php/art-n-soul-old-debates-about-photography) > - **That petition was also a fight over a category !** The most celebrated academic painter in France signed a public protest to keep the new medium out of the “Art” room. **It was a CPCWiki category poll with better moustaches. Nobody could have picked a more precise analogy.** > - **“Recording device vs synthesis machine” misses what the analogy was for.** It was never about the mechanism. It was about the claim: this makes a hard-won skill cheap, so appreciation of the skill will die. Photography made accurate rendering cheap. Value then moved to vision, composition and intent. That is reno08’s “skill moves up the stack” argument from #98, which Gemini skipped entirely. > - **Cherry-picking.** Gemini attacked the analogy from 1859 and ignored the one from 1983, “Real Programmers Don’t Use Pascal”. That one is about the same domain, the same decade and the same machines. It also ignored the register allocation → compilers → architecture argument. Picking the target that looks weakest while ignoring the strongest is just selection bias. > - **The ironic part.** Baudelaire’s actual prescription was that photography should be the “very humble servant” of the arts. **That is almost exactly the position several anti-AI posters took: “great tool, but not art”**. The analogy was predictive, not lazy. > > #### 2. The “Slippery-Slope Fallacy”: Gemini can’t name its own charge > > - **It names two different fallacies in one paragraph.** A slippery slope predicts a chain of consequences (“if we allow X, then Y, then Z”). A _reductio ad absurdum_ shows that a principle, applied consistently, leads somewhere absurd. A reductio is a **valid** form of argument. Gemini called reno08’s post one and then the other, and neither is an accusation if you get the definitions right. > - **The admin raised the line problem first.** Gryzor in 603 #4: “where do we draw the line?” He also called cpcitor’s Linux example an absurdum point but said “I do get it”. Then in #52 he wrote that there are no hard, naturally occurring lines and the line is “what we make of it”. So the “deliberate blurring” Gemini attributes to reno08 is in writing from the moderator. And “deliberately” is mind-reading, not argument. > - **Gemini’s own bright line falls apart.** It distinguishes ordinary tools from “systems that generate the assets or logic for you”. A compiler generates machine code. ConvImgCPC generates pixels from a photo. And Gryzor’s own form has a category for **“Port / Translation (AI converted code from another platform/language)”**. Translating code from one language to another is what a compiler does. By Gemini’s definition, a large share of the “AI” category is on the tool side. > - **“The other tools were easily accepted” is also false.** The demoscene argued for decades over converted versus hand-pixelled graphics, and compos still ban conversions. That fight was settled with **compo rules about the work** , not with a separate room for the people who did it. > > #### 3. “The Real Blind Spot” is a strawman > > - **reno08 never argued for “maximum corporate output”.** His stated argument was about game design, taste, artistic direction, and more people being able to create. He wrote, “If you love programming by hand (I do…), keep doing that.” Gemini answered an argument nobody in the thread made. > - **“The entire point of the retro scene is the human effort” is a false universal.** The thread itself contradicts it. RedBug (#87) says 99% of players don’t care about the technical side. AndyCadley (#14) describes people finally building the game they dreamed of as kids. Typhon (#79) wants to enjoy his free time rather than hand-write tooling. WacKEDmaN built an emulator. The scene is plural. Gemini took one faction’s motive and declared it the essence of the scene. > - **The limitation is still there.** A Z80 at 4 MHz with 64K doesn’t care who wrote the code. Typhon (#79) dropped smooth CRTC scrolling for flick-screen **even with AI help**. The constraint lives in the hardware, and AI hasn’t removed it. > - **“Silicon Valley disruption seminar” is an ad hominem by aesthetic.** It’s a vibe, not an argument. > > #### 4. It didn’t read the file > > A competent prosecutor had real material: the RETO translator error, the Bulverism, the Vespertino jab. Gemini found none of it. It charged reno08 with offences he didn’t commit while the actual evidence sat unread in the thread. That makes the analysis both unfair and incompetent, which is a rare combination for something posted “for giggles”. > > > _Hello, 911? I’d like to report an unsupervised Flash model._ ---
**Author:** ![reno08](https://forum.cpcwiki.eu/letter_avatar_proxy/v4/letter/r/48db29/32.png) [@reno08](https://forum.cpcwiki.eu/u/reno08)\ **Post date:** [October 11, 2026, 10:13pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/135 "2026-10-11T22:13:22Z")
And it doesn’t stop there … 😄 > ## The opposition’s fallacies, and where reno08 called it > > #### Rhino > > - **Anachronism.** In #61 he said any painter outside “Hyperrealism” had nothing to fear. Hyperrealism is a movement of the late 1960s and 1970s. In 1859 the dominant Salon style was academic realism, exactly the mimetic style photography threatened. Ingres signed the petition for that reason. > - **Unfalsifiability.** In #108: the skill AI needs today “is irrelevant… since it will likely soon no longer be necessary.” Any evidence from the present gets dismissed with speculation about the future. That isn’t an argument anyone can lose. > - **A circular stigma argument.** In #95, a lamer might lie about using AI, “which indicates these are ‘tools’ of a different nature.” The stigma creates the incentive to hide AI use. The hiding is then cited as proof the tool deserves the stigma. The argument supports itself. > - **Moving goalposts.** In #84, the important thing is “always to be honest about it.” After BSC (#94) showed nobody ever disclosed their other tools, #95 retreats to disclosing only if someone asks. > - **“The AI copied the code of the true creators” (#86)** is an assertion with no evidence. WacKEDmaN (#44) reports Claude refusing to copy emulator code and reimplementing it instead. CRTC scrolling techniques are documented publicly on CPCWiki itself, and reading public documentation is how every human coder learned them. > - **An uncharitable reading of Typhon.** Typhon’s “I have deliberately left out AI-assistance” (#79) meant left out of his tag proposal. Rhino read it as concealment (#81), and Typhon had to correct him (#82). > - _Credit where due:_ his counter-Bulverism in #108 was a clean hit. > > #### rexbeng > > - **“Never replaced painting, not even portrait painting.”** The portrait miniature trade largely collapsed within a generation of the daguerreotype, and many miniaturists became photographers or photo-colourists. > - **“It wasn’t photography, it was the paint tube.”** That’s a false dichotomy. Both mattered. Photography’s influence on Degas’s cropping, for example, is standard art history. > - **“Most artists didn’t care.”** He gives no evidence, and the 1862 petition is counter-evidence. > - **“This hobby will not evolve, it is artificially locked.”** The CPC scene’s 40-year history is one long evolution: cross-development, emulators, Arkos Tracker, cpctelera, ZX0, the M4 board, CRTC tricks discovered decades after release. The machine is frozen. The scene never was. > - **A self-contradiction.** In 603 #48 he wrote “we are painters and in this place photography is not allowed.” In #51 he denied that meant an exclusive club. A space where outsiders’ tool is not allowed is an exclusive club, so he has to pick one. > - **“Tech bros” (#37).** Ad hominem. > > #### Gryzor > > - **“The other tools were easily accepted” (#52)** is historically false. See the conversion debates above. > - **“The ad extremum arguments are silly”** (603 #53) sits awkwardly next to “that is an absurdum point, but I do get it” (603 #21). > - **“Tu quoque is a logical fallacy” (#118)** was half right. reno08’s #115 _opened_ with a tu quoque. Its closing move was a parity argument, though: machine-assisted expression for non-native speakers versus machine-assisted coding. That is not a fallacy. > - **#126** is an appeal to artificial authority, posted in breach of his own disclosure form. > > #### RETO\_AMSTRAD > > - **Reification.** In #45 and #114 he insists AI is a “SUBJECT” that takes decisions, and builds a policy on that metaphysics. WacKEDmaN (#46) dismantled it. > - **“We’re all going to end up speaking Arabic” (#119)** is a gratuitous non sequitur, and an ugly one. > - **The best argument in either thread is his, and reno08 never engaged with it.** In #43 he compares Olympic weightlifting with a mechanical shovel: same output, different purpose, so separate categories. That argument concedes the output can be equal and says the category is about what the activity is _for_. It is the one argument the “it’s just another tool” line doesn’t answer. > > #### The rest > > - **Executioner** calls it a plagiarising autocomplete with “Zero Intelligence” and then, two lines later, a great tool. He also made the translator mistake first, at #12, so reno08 has company. > - **slider271** says “I don’t think anyone rational disagrees” that AI was trained on stolen content. That’s poisoning the well. He gets credit for openly admitting he’s being hypocritical. > > #### Where reno08 called it > > 1. **Disclosure asymmetry.** Nobody ever had to disclose their other tools. BSC (#94) agreed, and Rhino conceded the point in #95. > 2. **80s cross-development.** Commercial studios did cross-develop for 8-bit machines on PCs (PDS and similar rigs). That was standard practice, not “cheating”. > 3. **The line problem.** The admin raised it in the very first post of the category thread, then conceded in #52 that no natural line exists. > 4. **The label becomes meaningless.** reno08 gave it two years. It took about forty minutes. After #98, AndyCadley (#99) and Typhon (#100), two active developers, announced they would declare _everything_ AI-assisted. A label applied to everything carries no information. > 5. **Moderation burden.** Gryzor (#121) is betting his mint 664 that more threads will need merging into this one, and he had already cleaned the category thread once (603 #81). > 6. **Credit is the real stake.** The opponents said so themselves. Rhino (#81) talked about “recognition… you don’t deserve”, and RETO (#114) said the scene “is nourished by recognition”. reno08 identified the right variable but got the explanation wrong. He diagnosed it as insecurity, when it is a legitimate norm about attribution. The interesting counter-model is the Erdős problems site: its maintainer stopped using credit-giving language for any result, human or AI, explicitly to disincentivise glory-chasing. In other words, it dealt with the status game by removing the status markers rather than adding labels. [wordpress](https://terrytao.wordpress.com/2026/10/06/changes-to-the-erdos-problems-web-site/) ---
**Author:** ![slider271](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/slider271/32/501_2.png) [@slider271](https://forum.cpcwiki.eu/u/slider271)\ **Post date:** [October 11, 2026, 10:38pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/136 "2026-10-11T22:38:53Z")
Jesus. Can we just let the AIs argue it out between themselves, while we get on with playing games on the CPC? 😛 ---
**Author:** ![slider271](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/slider271/32/501_2.png) [@slider271](https://forum.cpcwiki.eu/u/slider271)\ **Post date:** [October 11, 2026, 10:40pm UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/137 "2026-10-11T22:40:06Z")
> [@Anthony\_Flack](#): > > And that will affect the value of everything. Does it reduce my desire to code by hand…? Maybe? And things I have done, I figure looking back that people may just assume it was all made by AI anyway. And that’s the sad part for me. I hope you do still code on the CPC. ---
**Author:** ![WacKEDmaN](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/wackedman/32/99_2.png) [@WacKEDmaN](https://forum.cpcwiki.eu/u/WacKEDmaN)\ **Post date:** [October 12, 2026, 12:16am UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/138 "2026-10-12T00:16:27Z")
well this turned into a laugh n a half!.. there seems to be alot of misconceptions and pure BS being thrown around about AI…not just here but directly from the big AI mobs themselves… theres a hell of a lot of hype being spewed to generate interest and money into their currently ‘running at a loss’ business and on the other hand there is alot of hype and BS coming from individuals who haven’t even used AI for whatever reason… and only seen the slop… im not having a go at anyone… but there is alot of misconceptions happening…eg, LLMs take ALL information from the internet and have it stored in their database…that is just purely false, why would an agent try to pull info from the old cpcwiki STILL to this day if its all stored in its database? too right people should be worried when stuff like AGI is being touted as the next big thing…BUT that is all just pure hype as well… LLMs will never achieve AGI in the sense its being pushed…the underlying system of a LLM is still just pattern recognition, the storage of them patterns is whats different take old school Eliza chat bot, its doing pretty much the same as LLMs, it would be fed with a bunch of words and their relationships (hardcoded training)…then you’d ask it something and it would match words you used with words in its database… LLMs just have a bigger database of the words and relationships to each other… its ALWAYS been up to the individual which tools they use to complete their projects/tasks etc… heres some examples…would you code totally by hand in machine code (binary or hex)…or do you use an assembler to abstract away the tedious parts?.. do you source colors for your paint from nature or do you buy your paint from a shop?.. its exactly the same thing… LLMs are a tool and nothing else… use them or dont… there should no be any community pressure either way… sure alot of slop is going to be produced, but at the same time alot of amazing stuff can come out of it…which still requires time, effort, and a human running the show… people should really be learning to spot the hype and pure BS!..learn critical thinking… it will help a ton in the current climate …im not sayin my views are 100% correct either!.. but i really think this is all gone too far into nonsense, all over another tool that “can” make things easier… or can produce slop… there has always been both sides to everything… fire, for god sake, it can give warmth, or it can burn ya house down… are you never going to use it and bag on people who do, just coz it burned a house down?!.. cmon really?! ---
**Author:** ![WacKEDmaN](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/wackedman/32/99_2.png) [@WacKEDmaN](https://forum.cpcwiki.eu/u/WacKEDmaN)\ **Post date:** [October 12, 2026, 3:02am UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/139 "2026-10-12T03:02:19Z")
the memory issue solved? [![](https://forum.cpcwiki.eu/uploads/default/original/2X/7/77e2715eac9d37f5ca95a186c2f05a92d3333b0a.jpeg "DeepSeek Solved AI’s Memory Problem") ](https://www.youtube.com/watch?v=4fLVq6HJsKc) TL;DW: convert text to OCR images, compress images to 10th of original size, save memory keep info!.. can be used for training and context! ---
**Author:** ![lightforce6128](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/lightforce6128/32/32_2.png) [@lightforce6128](https://forum.cpcwiki.eu/u/lightforce6128)\ **Post date:** [October 12, 2026, 3:57am UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/140 "2026-10-12T03:57:55Z")
Such a “lively” discussion. So many arguments (in more than one meaning). I read and liked here and there, but there were some things I would like to comment (and test the forum at the same time how many links it can manage): > [@BSC](#): > > My impression is that it’s quite unusual to fully disclose all the tools (other than AI) that were used to develop a product. > [@WacKEDmaN](#): > > LLMs are a tool and nothing else… use them or dont… there should no be any community pressure either way… I have difficulties with calling an AI agent a tool. I see a qualitative difference e.g. between using assembler, a high(er) level language like BASIC or C, or a coding agent. The same applies to coding pixels manually, using modern Gimp or Photoshop, or using an image generating AI. I think, the verbs “to abstract” and “to delegate” describe the differences between these things. > [@Rhino](#): > > What we need to ask ourselves is where AI is taking us. This is the right question. Using an AI agent is not the end, it is still the beginning of this new era. The more prompts are written, the more AI (companies) will learn how to prompt automatically. The more games are converted or rewritten, the more AI (companies) will learn how our and other communities work internally. How will our world look like in ten years? > [@Anthony\_Flack](#): > > I think the automatic translation on the forum is neat, although I guess it has the potential to open up hitherto-untapped streams of inter-scene hostility as everyone can argue in their native language. Maybe it can also be of help: When people think, they are smashing each other, in fact the translation could silently alter or remove the worst accusations and moderate the conversation. Would anybody know? > [@Anthony\_Flack](#): > > People who previously lacked the patience and attention to make software can now create games without patience or attention. Other people who put the effort in will still be able to make worthwhile things, but we are going to be drowning in careless slop, I don’t doubt it. Because it can be turned out so quickly. Yes, this problem is already rising and probably will become much worse. The mechanisms used until now to promote good games and programs and hold back bad ones are getting at their limits. I think, this is one of the main points of this discussion: What is a feasible mechanism for the near future? > [@WacKEDmaN](#): > > but there is alot of misconceptions happening…eg, LLMs take ALL information from the internet and have it stored in their database…that is just purely false, why would an agent try to pull info from the old cpcwiki STILL to this day if its all stored in its database? For training an LLM, usually as much data as possible is used, because somehow it turned out that in this case it is quantity over quality to get better results. With this the LLM learns about words, meaning of words, relationship of words, classes of texts, reasoning patterns, etc. But after the training, it will not remember every detail correctly and hallucinate - the training process is like a lossy compression. A second layer is used afterwards: Automated prompting to guide the LLM and injecting detail information into the context window to reduce hallucinations. To do this, the surrounding infrastructure, not the LLM itself, needs to collect current information (from cache or from web), although it might have seen this before during the training. And I’m sure, the real process is even much more complex. But in a way: An LLM has stored a lossy compressed version of the whole internet in its database/weights. ---
**Author:** ![WacKEDmaN](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/wackedman/32/99_2.png) [@WacKEDmaN](https://forum.cpcwiki.eu/u/WacKEDmaN)\ **Post date:** [October 12, 2026, 4:30am UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/141 "2026-10-12T04:30:25Z")
> [@lightforce6128](#): > > But in a way: An LLM has stored a lossy compressed version of the whole internet in its database/weights. sorry but that is another misconception… the whole internet is not stored or scraped.. youd see way larger models if that was the case… most have been streamline and distill the models too, making them good in various subjects after they have been trained on actual language usage (verbal and machine, mathematical languages).. most of the real data isnt stored and only its relationships are…thats what weights are! weights are the real outputs of the LLM before its converted back into text..(eg the closest weight that corresponds to a token or word in a given context) the emulator im building with claude is from pure documentation and other freely available code that i gave it (it used code as documentation and wrote its own function that does the same job).. the only times its searched the net for information has been around expansion devices (eg MSX ports and v9990) which i didnt have docmented..all other times it would search my RAG database of documentation… now if it knew everything even if lossy..it wouldnt have had to search the net for the MSX and v9990 info right?! … if you look at the vid i posted, lossy means missing tokens here or there within the final text… which are then predicted back through the LLMs weights, An Agent is just a traditional peice of software with own functions etc that acts on the LLMs input and output via traditional functions.. ---
**Author:** ![Gryzor](https://forum.cpcwiki.eu/user_avatar/forum.cpcwiki.eu/gryzor/32/7_2.png) [@Gryzor](https://forum.cpcwiki.eu/u/Gryzor)\ **Post date:** [October 12, 2026, 5:53am UTC](https://forum.cpcwiki.eu/t/ai-mega-thread/743/142 "2026-10-12T05:53:51Z")
> [@Anthony\_Flack](#): > > I think the automatic translation on the forum is neat, although I guess it has the potential to open up hitherto-untapped streams of inter-scene hostility as everyone can argue in their native language. But it’s cool, it works just like Douglas Adams’ Babel fish. Auto-translation of Youtube videos is great too. To be frank when I enabled it I did so to accommodate the occasional native language post and the inevitable “English, please” replies. What I did _not_ anticipate was people beginning to write in their own language regularly. Which, ok, is not bad per se, if it helps people and allows them to post more, all the better. But, as you say, this can lead to problems, so people must be cognisant of the fact. > [@Anthony\_Flack](#): > > And I did wonder why not use an AI to migrate all the old forum posts to the new forum. There’s actually a mini discussion on it in the forum chat. Of course I’ve used AI to help me with the migration. However, even with the help of AI, it requires a TON of planning. It’s not just a matter of mapping user fields and post data - I’ve done this already, that’s the easy part. The trick is to find the logic differences between the two systems and work accordingly. There are many, many considerations I’ve come up with, and new ones appear all the time, and part of it has to do with the fact that I don’t know the new forum that well yet (it’s a whole different kind of _beast,_ much more complex than SMF). For instance, only yesterday it occurred to me that, "wait a minute, if I port all the old posts, I must make sure they are marked as _read_ otherwise every user will find themselves with **17.5k unread posts**! And, yeah, it sounds simple, but there are tons of such things to consider. There are nice discussions on SMF-\>Discourse migrations on the Discourse forum itself that do show some of the issues, and nobody has come up with a satisfactory solution; who am I to disagree with much more knowledgeable people? Also, relying on AI to do that for you is asking for trouble if you don’t completely understand what you’re doing. Who says that we don’t succeed in our migration, only to meet severe issues down the line? I’m no database engineer, so I can’t predict that. or make sure it doesn’t happen. Finally, there’s the small matter of available time. Even supposing I was wildly optimistic about the whole thing, it would still require large chunks of free time to attend to it, which I simply don’t have. …so there you have it. AI is a tool, but is not a all-in-one solution. [Previous page](https://forum.cpcwiki.eu/t/ai-mega-thread/743.md?page=6) [Next page](https://forum.cpcwiki.eu/t/ai-mega-thread/743.md?page=8)