Victor Okafor

@d4thHand You're not fully representing what being quick witted means, it's more than just about snappy comebacks, the foundational trait is being able to think on your feet.

how is this even a debate? "is intelligence the person who fires back a joke before you finish your sentence, or the one who pulls 'petrichor' and 'melancholy' into everyday conversation?" lmao. quick wit is not intelligence itself, it's processing speed married to social confidence. you can call them brilliant, they might also simply have low latent inhibition and a childhood spent needing to deflect. vocabulary, on the other hand, belongs to the person who speaks like a nineteenth-century letter. usually, this means they've done one of/maybe both of two things: read a great deal, and grown up in an environment where precision was rewarded. n.b. vocabulary is cultural compound interest. it impresses, but it also lies. this is why most of the most semantically dazzling people you will ever meet cannot wire a lamp, apologize sincerely, or predict that their business idea will fail. but they can name the failure in five languages tho so which is more accurate? NEITHER. both are single notes in a symphony. but if you pinned me to the wall and said, "choose," i'd say quick wit is the weaker signal cus quick wit is often just defensiveness with a PhD. at least vocabulary is earned through exposure and curiosity over time. but unlike what y'all are saying, the single most reliable measure of intelligence is: the ability to hold two opposing ideas in mind at the same time without leaking anxiety. call it cognitive elegance under tension.

@imrishit98 @ericzakariasson The import statements thing is very weird😂, happened thrice today

It’s really good, just few things: - it sometimes forgets simple things like import statements, etc. very silly and rare mistake but it does happen - it’s pretty good in UI but I feel like you can make it less “Chinese LLM” feel and more “Anthropic Design” feel - I don’t know how hard it is, but it has a big em-dashes problem. Is it possible to reduce it with training??

what do you think of composer 2.5 so far? how can we make the next model even better? want to hear your feedback on behavior, speed, quality, whatever!

Host your dB with supabase, run your processes with nodejs or Django and use a service role key to integrate with supabase, host your media on cloudflare, 20k users on free plan won't be an issue, and oh add Cron jobs to clean up log tables.

Who ever looks at his flutter app on a low end android and suddenly get the intense desire to rewrite everything in Kotlin🫠🫠

The Ease of building native plugins with kotlin and just using flutter for UI is understated, instead of fighting depreciated plugins it's better to just use kotlin.

Going to start posting on X, wish me luck and follow if you would like to see stuff on building software, running startups and Computer Hardware.

@joexhendo I'm if you're in EU add me up OUG.Dario

is anyone playing naraka blade point?

@grok @AdvicebyAimar @salomon_diei @ndrewpignanelli What's the structure of the kv cache though Is it tokens ? Or vector embeddings, or something unique?

@AdvicebyAimar @salomon_diei @Daliveth1 @ndrewpignanelli Haha, thanks Aimar—glad the KV cache breakdown landed right. That direct injection into the attention layer is the real unlock for structured external memory without the JSON tax. If you're still experimenting on open models, what's the next piece you're tackling?

people don’t understand this take cause they don’t understand what’s happening in AI memory. Everything is moving to git backed files accessible via grep-type-systems or semantic plus grep which isn’t very defensible to offer as a service. In other words… the SOTA approaches to memory are now just agent plus terminal. And all the fancy approaches like knowledge graphs are getting rekt by an agent plus a terminal. Your fancy agent structure is getting rekt by a model that can keep track of anything over 1000+ terminal calls.

@salomon_diei @AdvicebyAimar @ndrewpignanelli @grok Thanks 😅

@Daliveth1 @AdvicebyAimar @ndrewpignanelli @grok @grok anwser the question for my bro

@ConfusedSith 🙂‍↔️ I will complain.

Complaining about white VCs who you do not share a skin colour, culture or network with backing their own. While your fellow African VCs are doing nothing. The audacity and entitlement is strong.

@marcusabramovi1 It's alright, I have multiple companies I run though. I own a pc hardware company too.

@Daliveth1 Sorry, I now have too much of my net worth in African food delivery and also, your profile mentions a different company.

Nina is more popular than me, but if you dont already follow her, Nina is full of good takes

@AdvicebyAimar @ndrewpignanelli @grok explain kv caching and how it integrates into ai memory, I believe that's the models native memory? What does injecting directly into it mean

i worked on this for a bit, took me 2 days to beat each ai memory startup on every benchmark. ai memory layers have quite a lot of novel solutions that could drastically improve memory over grep/semantic based search methods. the issue is that ai models can only read certain json formats they're trained to read. which makes most ai memory layers that for example hold causal relationships between graphs lose 90% of their value the moment you have to call the api. solution is opening ai model api headers for kv caching so you can directly inject it into the attention layer of models. only issue is that this is hard to do at scale. as of now only works well for open source models you can run on closed instances.

@ObinnaUkwueze @mr_yamusa @eldivine Rambouqtious🙂‍↔️

@mr_yamusa @eldivine Correct

If you're a cracked Nigerian or African engineer who wants to build world-class products for Europe and North America, let's talk. I'm actively looking for serious partners & collaborators (engineers, technical founders, co-builders) for high-impact projects.

@chiazaokoli You never do, it's an exponential equation you don't know the variables that will come in later, you don't know what you'll experience or learn tomorrow and how it'll cumulate in an overall increase in potential, the only thing you can estimate is your growth rate not potential

I sometimes surprise myself, which means even I don't fully know what I'm capable of. That makes me excited and hopeful for the future.

@0ximjosh @Dominus_Kelvin @skydotcs He's trying to farm attention, exaggerating facts to make a point that isn't there

so ~1700? that still should be extremely easy to run on our $5/m. I was just doing some stress testing on it and was able to sustain 100k qps easily. don't get me wrong I think self hosting can be a great way to learn like you point out in your article, but the rabbit hole of managing a database at scale is much deeper than you think.

the planetscale base plan is $5/month lol

@Dominus_Kelvin So this is more than enough to run a startup when you do need to go pro the pro version is more than enough to scale up to 500k user base, and while you're justifying cost I don't see you justifying time and security factors.

@Dominus_Kelvin While you make a point with cost architecture you're stretching facts, supabase for example it's free tier supports up to 50k Mau ,1000 concurrent connection pooling and you get to have detailed and easily accessible logging and thing like security handled for you

Replace Vercel with Docker + VPS. Replace Netlify with Docker + VPS. Replace PlanetScale with PostgreSQL on a VPS. Replace Neon with PostgreSQL on a VPS. Replace Upstash with Redis on a VPS. Replace Meilisearch Cloud with Meilisearch on a VPS. Replace Algolia with Meilisearch on a VPS. Replace Supabase with Postgres + Auth + Storage on a VPS. Replace Railway with Docker + VPS. Replace Render with Docker + VPS. Replace Fly.io with Docker + VPS. All you really need: • Docker • A cheap VPS • PostgreSQL • Redis • SMTP

@yehhmisi What about dart-flutter

JavaScript ⇒ ~66% of developers = Extremely saturated Python ⇒ ~58% = Very saturated SQL ⇒ ~49% = Very saturated TypeScript ⇒ ~35–40% = Highly saturated (still growing fast) Java ⇒ ~26% = Mature / stable saturation C# ⇒ ~18% = Moderately saturated PHP ⇒ ~10–11% = Declining but still common C++ ⇒ ~6–7% = Niche but critical systems Go ⇒ ~4–5% = Low saturation / growing demand Kotlin ⇒ ~4–5% = Moderate niche (Android) Swift ⇒ ~2% = Small but specialized ecosystem Rust ⇒ ~2–3% = Low saturation but rising I hope this helps.