humbleferret

Nice writeup! I imagine these results change as harnesses are updated, so you'd need to frequently rereview.

I'd love to see a tiny, reproducible benchmark repo that anyone can drop on their own hardware and then run against all harnesses at once to compare the per turn prefix token count, time to the first token, experienced tokens/sec (and prefill), cache reuse % and a pass rate on a deterministic set of small tasks. I think it could also be useful to have some way to share results and hardware for others to compare.

Great film!

This piece hits on the idea you can cultivate luck. It's about increasing your 'Luck Surface Area' (L = D x T: Luck = Doing x Telling).

My notes/ points that stood out for me:

- Actively seek diverse social, intellectual, and professional interactions. Each one expands your surface area for serendipity. TAKE MORE SHOTS!

- Don't go into interactions with an agenda. Real interest opens doors unexpected.

- Always aim for excellence. Even in small tasks. People notice this. A strong reputation pulls in opportunity.

- Give before you take. Offer attention, time, or resources without expecting immediate return. That builds high trust relationships.

- Share your authentic self. Put your unique thoughts out there. It's a magnet for those who truly resonate with you. Don't fake it.

- Host stuff. Simple, low pressure events work best. Taking the initiative to create connections is often more effective than just showing up.

I think some of the sources like the Reviews.org survey, JAMA analysis and Sleep Medicine study have flaws and are largely correlational, not causal. The Reviews.org data is self reported, not peer reviewed or randomised, and probably has sampling bias. JAMA's analysis is observational, so it doesn't prove causation. The Sleep Medicine study uses a tiny sample, short duration, and likely has self selection bias.

Regardless, we all sense that excessive smartphone use ties to lower mood and higher depression risks. There's already an influx of people experimenting with timers, dumbphones, or just switching their phone off, but app builders will keep chasing our attention no matter what. It's their core business. As the patchy research (in both quantity and quality) shows, real change won't come until we as a society dig deeper into the problem.

key points that stood out to me:

- Smartphones are built to addict via dopamine from notifications and scrolling. Americans check theirs 144 times a day on average. Strong links to anxiety, depression, poor sleep, and weaker focus.

- Short breaks (24 hours to a week) can boost focus and mood. Limiting non work screen time to under two hours daily correlates with lower depression.

You can't completely outsource defence.

International cooperation is important, but over reliance on a single foreign partner, even a close ally causes risks. It leads to a situation where a country possesses shiny new military hardware but lacks the logistical support, sovereign weapons integration, skilled personnel, and overall control to use it effectively and autonomously.

I imagine the UK's experience with the F35 is a major driver in it's decision to partner with Japan and Italy on the next gen Tempest fighter, so they can regain sovereign control over its future air combat capabilities.

However, the problem extends beyond the F35. Although the UK has increased its defence budget in real terms, the extra funding has been outstripped by soaring equipment costs (Boeing mark up the costs of soap dispensers by 7,943%!), the legacy of decades of under investment, and the spiralling expense of strategic programmes like the nuclear deterrent. The budget for conventional forces does just not exist.

What a great post.

Some points that stood out to me:

- Progress is iterative and driven by a seemingly bottom up, meritocratic approach. Not a top down master plan. Essentially, good ideas can come from anywhere and leaders are promoted based on execution and quality of ideas, not political skill.

- People seem empowered to build things without asking permission there, which seems like it leads to multiple parallel projects with the promising ones gaining resources.

- People there have good intentions. Despite public criticism, they are genuinely trying to do the right thing and navigate the immense responsibility they hold.

- Product is deeply influenced by public sentiment, or more bluntly, the company "runs on twitter vibes."

- The sheer cost of GPUs changes everything. It is the single factor shaping financial and engineering priorities. The expense for computing power is so immense that it makes almost every other infrastructure cost a "rounding error."

- I liked the take of the path to AGI being framed as a three horse race between OpenAI (consumer product DNA), Anthropic (business/enterprise DNA), and Google (infrastructure/data DNA), with each organisation's unique culture shaping its approach to AGI.

Barbarian Days: A Surfing Life by William Finnegan

No better way to be reminded there is a big difference between someone that surfs for fun and someone who is a surfer.

Surfing parts aside, this book taught me to appreciate the different parts of your life around your 'obsession' as you get older.

I think it's circumstantial and dependent on individual situations. I personally think you should optimise for equity as much as you can. If you wanted to optimise just for cash, you could have found a better paying job most of the time.

From the small sample size of my friends in SF/NYC and are well funded by a top-tier VC, most take lower salaries for more equity. The majority have the same salary as their seed round with small bumps along the way, but nothing crazy. A couple have reduced their salaries to extend runway without diluting themselves further.

The only exception seems to be those with families and dependents.

'Prompt Injection' is everywhere. While it's cheeky, you can't help but feel it's a fair countermeasure. If the academic publishing world is so overwhelmed that reviewers are resorting to AI, maybe the system itself is the real problem.

The only people affected are the lazy reviewers who weren't doing their job properly to begin with.

A system that matches by name rather than a unique ID is wild. It could easily identify the wrong people.

Great work!

Are there any particular niches or applications where community-sourced splats are especially well-suited? I see them mentioned on HN from time to time, but I'm curious about their primary use case at the moment.

I was curious so looked at a few power wheelchairs...It's wild how expensive they are, especially considering the advancements in electric mobility tech elsewhere. You'd think they'd share some components with e-scooters, e-bikes, or even electric cars – motors, batteries, controllers.

Are the powertrains and control systems in power wheelchairs really that specialised? Or is it another case of the medical device markup and regulatory hurdles driving up costs?

Blockchains are not intended to be a universal solution, they excel in different ways, particularly enabling trustless and censorship-resistant value transfer. Which, for many new chains, is still their most used function.