🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦

Video game company; creating the app/game C++ engine "neoGFX". Part of team that invented, in 1999, the smartphone (Steve Jobs didn't). RTs aren't endorsements.

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

Claude (AI) was very useful in helping me fix a bug in my renderer tonight - a problem I have been grappling with all week. AI is much better than Google search IMHO. #AI #compsci #coding #gamedev #cpp #neoGFX

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

https://t.co/7Q2NkFPJGH

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

I guess we can add AI-induced brain atrophy to the list of things that are ending civilisation. #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

@ArthurMacwaters Aesthetics? The rockets sill look like giant dildos.

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

@rickygervais Most important question: will Brian Gittins feature?

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

I think the best way to use agentic AI for code generation is a TDD approach: the human needs to curate the tests which, as a nice side effect, can also be used to direct architectural design.
Verification is the most important aspect of software engineering, always has been, always will be, even in this new AI industrial revolution.

#AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

https://t.co/Tjf1njgZ2d

#AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

AI is amazing but will the hard problem of hallucination ever be solved?

#AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

Early this year I used ChatGPT 5.2 to create an alternative to std::chrono::steady_clock/high_resolution_clock: it was slower than latest VS2026 C++ stdlib; tonight I got Claude Opus 4.8 to improve it, it is now over twice as fast as VS2026 C++ stdlib and hopefully behaves in a sane manner after Windows suspend/resume (unlike the VS2026 C++ stdlib). #cpp #coding #gamedev #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

#AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

https://t.co/LAMIY05SOU

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

couch potatoes like to lounge in a comfy couch, AI is turning us all into couch potatoes. #IRC #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

I can confirm that Claude's memory (context window) works across model variants (Opus <-> Fable). #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

I am currently drunk having consumed a whole bottle of chianti .. so there is also that -- LLMs don't get drunk unless you ask them to pretend to. #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

Claude Fable 5 knows C++ better than me. Does that make Claude Fable 5 a better C++ programmer than me? I genuinely have no idea -- I am getting old so there is that. Interesting times indeed. #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

Although I guess for the major frontier models there is deliberate curation of the training data .. higher weightings for high quality data sources.

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

On LLMs. The fundamental problem with these clever stochastic parrots is that the bulk of their training data is the low quality parts of the Internet that causes them to have problems with hallucination when confronted with the more obscure/unique aspects of human knowledge. #AI #compsci

[🇺🇦ᓚᘏᗢ 🕹🎮i42 Software🎮🕹 ᓚᘏᗢ🇺🇦 @i42Software]

https://t.co/mpZlZNdl28 #AI