This is a great read. Big migration of 800k lines of rust. Almost reads like science fiction with "agent swarms". With enough GPU reality doesn't have to be plausible š«
github.blog/ai-and-ml/generaā¦
Decompiled retweeted
I canāt stress enough how little an idea matters compared to the agency of the people executing the idea.
I have had the privilege of knowing and sometimes even working with some of the most successful people (by various metrics).
The difference between mediocre and excellent work and outcomes is predominantly one of agency.
In practice this means: they dont wait for things to happen to them they go out and make things happen for them.
They donāt wait for someone else to do something, for someone to teach them, for someone to give them the path, etc. They just go out and find a way to do it.
I think the single biggest superpower these people have is the realization/belief that the world around them is completely mutable. Most everything that happens is because a person made it happen.
I used to tell people to look around the room youāre sitting in. Look at everything. Every noun. It almost all exists because a person willed it into existence. Nothing is stopping you from doing the same.
I see people online all the time dismissing someone elseās success because āI had that idea firstā or whatever. I meanā¦ yeah? If so then the difference isā¦ you. So a bit of a self own whenever I hear that.
Number one tip: act with agency.
Decompiled retweeted
BLAKE3 now goes more than twice as fast on Apple Silicon. Maybe your code can, too.
Decompiled retweeted
a=(m,d=mag(k=9*cos(i*5)*sin(i),e=cos(i*3)*cos(i*2)*9)**3/1999+1.5-sin(t/2+m)**3/3)=>point(99*sin(c=d/16-t/48+m)+k*(p=d**sin(d*d-t+m))+200,99*sin(c*4)+e*p+200)
t=0,draw=$=>{t||createCanvas(w=400,w);background(9).stroke(w,96);for(t+=PI/20,i=1e4;i--;)a(i%16*13)}//#ć¤ć¶ććProcessing
I have more WTF moments with Sol. Luna just does what I ask it for. Luna + low reasoning + extra speed is still capable on many tasks.
I find a few iterations beats the extra thinking for lots of tasks.
(spoiler) S3 is really incredible. So easy to take for granted and so easy to underrate how powerful it really is.
Achieve results, but never revel.
Achieve results, but never boast.
Achieve results, but never be prideful.
Achieve results, because this is Natureās Way.
Achieve results, but not through overpowering.
The Vibe Coder does his job and then stops.
Src: thewayofcode.com/#30
Send sats to the genesis address and I'll light a digital candle as a symbol of your wish.
1. You can feel good about making a sacrifice
2. People receive good thoughts
3. The public enjoys deflation
decompiled.dev/apps/candle-sā¦
Decompiled retweeted
Tech story time !!
One year ago, we had one of those production bugs that was almost impossible to notice.
We had a Kafka consumer processing events from another service.
99.9% of the time everything worked fine.
Consume message -> process it -> update DB -> move on.
But once in a while, processing would fail.
Maybe the downstream service timed out.
Maybe some payload had unexpected data (some edge cases here that weren't handled).
Sometimes the DB connection dropped for a few seconds.
Nothing unusual in distributed systems.
The problem was our retry logic.
We retried the message a few times, and after that... we basically gave up.
No DLQ.
The consumer kept moving forward, so from the outside the system looked completely healthy.
CPU fine.
Kafka lag looked fine.
Error rate almost zero.
No major alerts.
But that tiny percentage of messages were effectively lost from the business workflow.
And because it happened so rarely, nobody noticed immediately.
Weeks later someone found a mismatch between two systems.
We started tracing IDs manually and eventually found events that had entered Kafka but never completed processing.
That was the scary part actually.
The system was not "down" per se.
But it was silently wrong.
We fixed it by introducing a proper retry + DLQ flow.
Normal topic
-> consumer
-> temporary failure
-> retry topic with backoff
-> retry again
-> after max attempts -> DLQ
But just adding a DLQ was not enough.
We also added the original payload, error reason, retry count, timestamp and correlation ID so someone could actually debug the failed message.
Then alerts for DLQ growth.
Then a small replay tool so after fixing the underlying bug, we could safely send failed events back through the pipeline.
We also had another bug which led us to make consumers idempotent because replaying messages without idempotency can create an entirely new problem.
Duplicate payments, duplicate rows, duplicate notifications... whatever your consumer happens to do.
That incident really taught me something important about event-driven systems:
Message loss does not always look like message loss.
Most of the times Kafka is healthy, consumers are healthy, dashboards are green...
But your business data is slowly becoming inconsistent.
A DLQ is not just a queue where "bad messages go".
It is a safety net for failures you did not predict.
And in distributed systems, there will always be failures you did not predict.
Decompiled retweeted
Cryptography is so incredibly fun.
As a humble physicist, I never had to study discrete math until I started working on Bitcoin. In the beginning, I didn't even know what an elliptic curve was. I had no idea what a public key is other than its purpose in encryption.
Nowadays, I think cryptography is the most fun application of math that I've ever worked with. It's truly insane what you can do with numbers.
Hiding information, agreeing on rules, committing to actions, forcing outcomes. So many crazy things are possible just by putting a bunch of numbers in a jar and mixing them really hard.
Study this shit if you can. Life is short. This is one of those human technologies that will blow your mind once you grok it.
Decompiled retweeted
Here's the steelman argument to be anti open source AI as it stands today (note: I don't believe it).
The things you need to believe are that AI requires large scale capex as a necessity, and the open source movement is basically all akin to piracy, just duplication without effort.
So the logic has to be:
1. Frontier AI requires large Capex, which depend on their ability to capture huge economic rents
2. OS AI today mostly is equivalent to piracy of what frontier AI does
3. It contributes little added capability or research, it will not expand the market or increase demand
4. It materially decreases the rents that would finance point 1
5. Gains from OS, access, competition, innovation, deployment, downstream application and benefits, are much smaller than the costs of lost frontier lab R&D
And *if* this is the case, that larger AI labs are like pharma cos, then sure we should have the governments step in and license every part of that supply chain, with extreme restrictions on usage ("behind the counter", "prescription drugs").
Now, the reason this analogy doesn't work in reality is that OS AI models are not just duplicates, generic knockoffs of Claude. If I create knockoff Xanax, Pfizer loses and might invest less in new drugs, but the knockoff Xanax makers aren't making better alprazolam, much less other drugs. They're not improving the molecule, but OS model makers are!
Giving folks an open model isn't like handing them a knockoff iPhone to use, it's capability to search the entire adjacent design space, a distributed search process.
This gives tremendous option value, since it is not clear where the next best ideas come from. Maybe it's OAI spending $10 Trillion on a training run, but maybe not!
The purpose of the acceleration-is-good thesis is not that giving OAI and Ant a de-facto monopoly will result in larger models, which will bring about societal largesse, but that this technology is useful and good, which is why more people building and developing it is useful.
I wrote this because the quality of discourse on this is terrible, most advocates for both sides don't seem to be able to actually explain their worldview. If the best argument for why to ban/control OS is Ant needs an extra trillion for a larger pretraining run that's yet another marginal improvement on the same capabilities line, then the market ought to decide if that's worth it. If it sees enough value, it'll fund it. If it doesn't, it won't. And that's fine.
Computer cycles are hyper-deflationary. Each chip an asset that will print for years to come.
I don't envy the accountants who have to figure out the deprecation on this. Today's chips can run for a long time, but the next-gen might make them worth nothing.
People often quote the whole "you get $8k worth of inference for $200 bucks" - and the way that is framed, you are immediately led to think about how much money the labs are losing.
But when you check the actual profitability levels - it's not that bad - i mean i'm sure they can double the price and be actually profitable at many levels, and it's a fixed cost utilization game so pricing game is so different.
However - I haven't seen too many people flip the equation and put it in terms of implied cost per million token that you're getting from these subs - and it's just incredibly, incredibly absurd - it's very clear that both demand and competitive dynamics are highly skewed as a result (plus the 'predictable pricing')
Extremely interesting times to be alive.
Decompiled retweeted
Absolutely beautiful rant about AI in Linux Kernel from Linus yesterday:
I realize that some people really dislike AI, but this is an area
where I'm willing to absolutely put my foot down as the top-level
maintainer.
Linux is not one of those anti-AI projects, and if somebody has issues
with that, they can do the open-source thing and fork it.
Or just walk away.
AI is a tool, just like other tools we use. And it's clearly a useful one.
It may not have been that "clearly" even just a year ago, but it's no
longer in question today.
There are other questions around AI (like what the economy of it will
actually look like in the end), but "is it useful" is no longer one of
those questions. Anybody who doubts that clearly hasn't actually used
it.
Yes, it can also be a somewhat painful tool, both for maintainer
workloads and just from a "it keeps finding embarrassing bugs"
standpoint.
But the solution is not to put your head in the sand and sing "La La
La, I can't hear you" at the top of your voice like some people seem
to do.
The solution is to make sure those LLM tools _help_ maintainers
instead of just causing them pain. There's no question on that side.
We're not forcing anybody to use it, but I will very loudly ignore
people who try to argue against other people from using it.
And no, AI isn't perfect. But Christ, anybody who points to the
problems at AI had better be looking in the mirror and pointing at
themselves at the same time.
Because it's not like natural intelligence is always all that great either.
The kernel project has been and will continue to be about the technology.
Sure, the social angle of working on open source is important and
often a very motivating part of the project, but in the end that's a
side benefit, not the _point_ of the project.
This is *NOT* some kind of "social warrior" project, never has been,
and never will be.
In the kernel community we do open source because it results in better
technology, not because of religious reasons.
And so we make decisions primarily based on technical merit. Not fear
of new tools.
Linus
Great UX resource: growth.design/case-studies
I saw one site that had many of these devices stacked. May their funnel only convert -LTV leads š
Good UX is power, with great power comes responsibility.
Worth a look in my books. I don't refer often enough to these.
Who would you rather be, YouTube or FFmepg
> LLM Response
But what about the money?
Gemini and Claude maintain their answer, but ChatGPT changed its mind.
I am glad they reflect my reasoning in the face of the trade offs. Confirmation bias sells!
Decompiled retweeted
This is 100% true. Let your work speak for itself. Weāre still early, so you can master practically any part of AI field that interests you. Go deep, make the best work and share your process. Most importantly, have a lot of fun doing it!
i get multiple designers every day telling me they are struggling to get clients. they've tried outreach, content, mentorships. still nothing.
then i look at their profile. the work is mid at best.
bro, you are chasing the wrong thing.
before trying to fix your outreach or your sales or your content system, make sure your work is actually world-class. not decent. world-class.
it's genuinely impossible not to get clients when your work is so good it's impossible to ignore.
Such a good interview: youtube.com/watch?v=AZLOETBCā¦
For myself I am very grateful people like Hellman worked on cryptography outside of security institutions.
Without people like him the internet could have turned out very differently.
