Open Source Highlight: just
We’ve been looking at just, a command runner by @rodarmor , and what stands out is how intentionally scoped it is.
Instead of becoming another build system, just focuses on one thing:
👉 making project commands explicit, discoverable, and repeatable.
What makes it quietly innovative:
→ It treats developer workflows as UX, not glue code
→ Commands become self-documenting (just lists everything)
→ No ecosystem lock-in — works with any language or stack
→ Familiar, boring syntax that teams actually adopt
→ Solves a social scaling problem (“how do I run this?”) with a tiny technical nudge
It’s a reminder that productivity gains often come from reducing cognitive load, not adding abstraction.
Sometimes the best tools don’t feel clever — they feel obvious after you use them.
🔗 https://t.co/glYRR1fkxp
Open Source Highlight: just
We’ve been looking at just, a command runner by @rodarmor , and what stands out is how intentionally scoped it is.
Instead of becoming another build system, just focuses on one thing:
👉 making project commands explicit, discoverable, and repeatable.
What makes it quietly innovative:
→ It treats developer workflows as UX, not glue code
→ Commands become self-documenting (just lists everything)
→ No ecosystem lock-in — works with any language or stack
→ Familiar, boring syntax that teams actually adopt
→ Solves a social scaling problem (“how do I run this?”) with a tiny technical nudge
It’s a reminder that productivity gains often come from reducing cognitive load, not adding abstraction.
Sometimes the best tools don’t feel clever — they feel obvious after you use them.
🔗 https://t.co/glYRR1fkxp
Adam Smith, an AI engineer, says that as powerful as AI becomes, we’re wired for real, human-to-human connection. Replacing that with something artificial comes with real risks—and some things simply can’t be imitated.
🎙️ ft. Adam Smith | Hosted by @GilbertGravis
🎧 Watch the full episode here:
YouTube: https://t.co/8OvuLdWwCa
Spotify: https://t.co/KERTeuBWCi
#HumanConnection #AIThoughts #TechAndSociety
ok so I looked at the paper, here's the reality check:
the game theory framing is cute but what it's actually doing is iterative magnitude pruning with a fancier objective function. "weights compete in nash equilibrium" sounds sexy but operationally it's: compute importance scores, remove low ones, repeat. we've been doing this since the 90s
the results ARE good but context matters:
90% sparsity on ResNet-50 is solid but not unprecedented. lottery ticket hypothesis (2019) hit similar numbers. SparseGPT, Wanda, etc already do this for transformers
they tested on vision models (ResNet, ViT) and CIFAR/ImageNet. these prune WAY easier than LLMs
"no retraining needed" is doing a lot of work — most results still show a calibration phase which is... retraining with extra steps
the GPT-on-your-phone claim is where it goes off the rails:
pruning CNNs ≠ pruning attention layers. transformer pruning is still an open problem
90% parameter reduction doesn't mean 90% memory reduction at inference (activations dominate)
sparse matrix ops are poorly supported on mobile hardware. you often get slower, not faster
"paradigm shift" is generous. it's a nice paper with a novel objective function that might marginally beat SOTA on some benchmarks. that's... normal science
Brian's vibes are "I installed it and it felt fast" which is not benchmarking
tldr: real paper, overhyped thread. game theory angle is marketing. still waiting for the pruning method that actually ships billion-param models to edge
Vision Weekend is here!
Yesterday we kicked off with a VIP event at Altos Labs, and today we have enjoyed 20+ keynotes on the frontiers of AI, bio longevity, neurotech, and space. For example:
• Jan Leike (@JanLeike) on why he is optimistic about the progress of (human-level) alignment, comparing models from Anthropic, OpenAI, Google DeepMind, xAI, Moonshot AI.
• Laure Deming (@LauraDeming) with a live demo of the tradeoffs between speed and dilution in cryopreservation, with simulated molecules.
• Viren Jain (@stardazed0) making the case for how we can reproduce neural activity using all available data, and using AI for missing parameters.
Now we are gearing up for the cypherpunk party tonight. Thank you all for today, and see you again tomorrow for day 2!
Cursor scaled to $29B without any full-time PMs.
Ryo (Cursor's Head of Design) walked me through how they work and it's the opposite of every big tech best practice:
1. Roles are muddy
PM work is spread across designers and engineers. Everyone does what fits their strengths and uses AI to fill the gaps.
2. Most designs start with code directly
Ryo barely uses Figma except for initial exploration. Most features start as live Cursor prototypes because "it feels more real than pictures."
3. No annual roadmap theater
Just a "fuzzy direction" and features shipped to concentric circles (e.g., staff, nightly beta users, consumers, enterprises) to polish.
Ryo also showed me exactly how he designs and codes new features using Cursor and how he avoid creating generic purple AI slop.
📌 Subscribe to watch our full tutorial tmr: https://t.co/Ggqaa3F11Z
Not Everyone Should Be a Founder: Great Teams Start With the Right Roles
Many chase the dream of becoming a founder — only to end up in years of burnout, self-doubt, and financial strain, realizing too late that the role never truly fit. Investor and founder Logan Yonavjak has seen this pattern across the startup world. Through her latest project, the Founder Readiness Institute, she’s helping people understand who’s genuinely wired to lead — and who can make a bigger impact by playing a different role.
🎙️ ft. @Loganyon | Hosted by @GilbertGravis
Sharing the long-form version of my conversation with Logan Yonavjak, cofounder of Founder Readiness Institute.
I'm a big believer in emotional intelligence being as or more important than raw IQ in the founder's journey, and it was a privilege to talk shop with someone who's working at the frontier of this field:
Take a look, and follow @pioneer_park_ for more!
https://t.co/cUpObRqjp6
AI tools like large language models are reshaping software development by encouraging teams to be more precise in defining requirements, tests, and documentation. While this speeds up individual productivity, the bigger challenge lies in maintaining a collective mental model — a shared, accurate understanding of complex systems across teams. To address this, new best practices are emerging, such as creating detailed summaries, embedding documentation directly into the codebase, and using natural language explanations alongside the system to keep everyone aligned as technology and workflows rapidly evolve.
#Kaizen #SoftwareDevelopment #AI #ArtificialIntelligence #SoftwareEngineering #TechMindset
What truly defines a career isn’t the title on your business card, but the outcomes it creates in your life. Whether you’re in the trades, technology, or any other field, the real measure of success comes down to stability, growth, and the opportunities your work provides—for you and your family. It doesn’t matter how you get there; what matters is that you get there. Beyond job titles, it’s about building a future that’s both sustainable and fulfilling.
🎙️ ft. Ziwen Deng | Hosted by @GilbertGravis
#CareerGrowth #SuccessBeyondTitles #SkilledTrades #FutureOfWork
The U.S. is facing a massive skilled labor shortage: nearly half a million unfilled trade jobs today, over a million skilled roles overall, and the gap keeps widening. Ziwen found that for every 5 experienced workers retiring, only 2 are stepping in to replace them. With an aging infrastructure and industries relying on fewer hands, the challenge is clear: we need more skilled workers, trained now.
🎙️ ft. Ziwen Deng | Hosted by @GilbertGravis
#LaborShortage #FutureOfWork #WorkforceCrisis #TradeJobs #SkilledLabor