Pinned Tweet
“If you want to build a ship, don’t drum up the people to gather wood, divide the work, and give orders. Instead, teach them to yearn for the vast and endless sea.” - Antoine de Saint-Exupéry – at Islington, London
models score 100% on AIME 2025. they hit 70%+ on professional benchmarks. yet businesses struggle to extract value.
the benchmark performance says genius. the P&L says otherwise.
research teams have entire divisions creating RL training environments specifically designed to boost scores. they treat tests like competitive programming practice. the model grinds 10,000 hours until every proof technique is automatic.
then it fails to fix a simple bug without introducing two new ones.
the economic puzzle: impressive numbers, underwhelming business impact. the sample efficiency gap tells you everything.
1
85
Note taking apps have come a long way since I worked on my own version way back in the dinosaur age of 2019. However, there's still a major issue around the directionality of the summaries and notes.
Almost always gets the name of the speaker wrong and thus messes up the whole formatting of the notes and the content, which I have to manually fix almost 70% to 80% of the time. I feel like there's a relatively easy way to fix this Pareto principle-wise. For example, if I say somebody's name, e.g. John, can you do this for me? The next person that speaks should be tagged as John, and thus that is the name of that speaker, and further notes in terms of the diarization should be handled as such.
Very straightforward fix, probably doesn't work for a number of reasons, but at least something has to give.
65
AI learned language from text alone. No eyes, no hands, no body.
Now imagine what happens when you give it context about your users. Their behavior. Their goals. Their exact moment of friction.
That's when AI stops being a chatbot and becomes infrastructure.
1
1
46
imagine ordering coffee at a shop where the barista forgets you every single visit.
you explain your order. every. single. time.
now imagine a barista who remembers you hate oat milk, knows you always add an extra shot on Mondays, adapts to your preferences.
that's the jump from stateless AI to memory-enabled agents.
SaaS products with memory don't just respond. they learn user patterns, recall past interactions, compound value with every session.
your users won't tolerate the forgetful barista once they've experienced the one who remembers.
2
3
61
chuka 🥣 retweeted
I'm incredibly excited to announce that @paid_ai has raised a $21M seed round led by @lightspeedvp with support from @fuse_vc and @eqtventures .
Since starting Paid in January, we've been working flat out to help our customers focus on building amazing products while Paid handles the complexity of monetisation. Our platform helps teams demonstrate their agent's business value, understand their costs, configure bespoke deals using our pricing configuration agent, and so much more.
I want to thank the team for all their hard work to get us to this point, and our amazing customers who we love working with.
We have big ambitions. If this sounds like your sort of thing, come join us!
chuka 🥣 retweeted
The Arch website received a nomination and special kudos in the @cssdesignawards submitted by 3d designer Johannes Held, who worked on the design alongside @lukaskmoth
First class UX for our visitors and users is always top of mind for us and we appreciate the recognition! Major updates in-app coming soon 👀
best place to go for reading the latest ML papers: github.com/dair-ai/ML-Papers…
43
chuka 🥣 retweeted
Excited to finally share what our team has been cooking at @phantom. Trading perps is about to get easier, faster, and much more fun. We're just getting started.
Introducing: Phantom Perps 👻 ♾️
Go long or short in just a few taps.
100+ markets. Up to 40x leverage. All in your pocket.
Powered by @HyperliquidX
right now i'm at my brokest but weirdly at my happiest. all because i get to do exactly what i want to. build the future as i see it
1
85










