Piotr Sarnacki
7.8K posts
Piotr Sarnacki
@drogus
I'm a software developer and an SRE https://t.co/PkEVyLbMrq
Berlin, Germany 参加日 Mayıs 2007
704 フォロー中1.3K フォロワー
@GergelyOrosz 1. sure
2. increased competition among engineers and higher expectations will stop you from doing this
In my reality 2x productivity is 100% true.
With the latest models and whatever new models are being released in next 6 months 2x is going to be the baseline imo
English
There is massive irony in how AI coding tools are starting to become TOO expensive for many enterprises - after eg Anthropic removed subsidizing AI subscriptions.
We might go from "everyone use AI for everything!" to "you have $300/month AI budget; use your brain for the rest."
English
@VitorAlcobia @but_cyclists @javiermalagon First maybe check who pays for all the car infrastructure and car related problems
English
@javiermalagon And who will pay for this????
No car, no money coming in, no insurance to pay, no road tax to pay, no petrol tax to pay, no mechanics needed, no parking tickets, no speeding tickets, all economy lost, so who pays for these ? 100% not the cyclists.
English
@v_oycieh @bandyci_drogowi Czyli teraz wyprzedzanie innego pojzdu to blokowanie pasa ruchu 😆
Polski
@bandyci_drogowi Na prawy pas by ustalić przejazd szybszym pojazdom. Nikt na drodze (prócz Policji) nie ma uprawnień by decydować kto ile ma jechać blokując ruch. Oczywiście pojazd z tyłu na filmiku mandat i psychiatryk. Ale na tego z kamerką też jest paragraf o blokowaniu ruchu 2/2
Polski
@but_cyclists It’s actually quite funny cause she is talking about the new bike path that requires the drivers to give way to bicycles 😆
English
Plot twist: the news report is actually about how SUV owners are wonderful people
English
POV: you try to do a news report from a bicycle but an SUV owner has other ideas
English
@AwareFelis @kanavtwt Yes, because as we all know, ancient hardware gets frequent updates with newest kernals
English
@kanavtwt This will prevent old computers from running Linux.
And by old computers I mean specialized hardware that don't get replaced as often as consumer hardware, like MRI Scanners or ATMs.
English
After recent Claude vs OpenClaw weirdness, I'm wondering if this is on purpose, response from Claude when asking about OpenClaw: I don't recognize "OpenClaw" — could be a typo.
English
The problem I have with “just give an AI agent a good feedback loop” is youcan’t account for everything. How do you write tests that disallow security vulnerabilities?
English
@corsaren I clicked the tweet and I thought "I need to bookmark this" only to realize I've already bookmarked it 😆
English
To the other 12,000 people that, like me, have had this bookmarked since June but never actually read it, I have an unfortunate announcement: it is very good. Read it.
Fernando 🌺🌌@zetalyrae
I wrote 8k words on how to manage ADHD
English
@Boenau interesting thing about cyclists...
No friends, no passengers, no kids, no shopping, no luggage, no flat-pack furniture, nothing other than themselves.
Rather selfish of them to hog half the road don;t you think?
English
"Residents won't use bikes or scooters."
"Commuters won't use bikes or scooters."
"Tourists won't use bikes or scooters."
And then something magical happens when you create a network of bike lanes.
👀
English
@GergelyOrosz The question is also how much time he (or other people involved) spent on managing the agents. A week of work sounds like the saved $5k
English
Congrats on having fun and building a vibe coded WYSIWYG editor
Patiently waiting for when they’ll realize things like permissions, tagging, backups+disaster recovery, search, exports, tables, mobile app, integrations w Slack+Linear+others need to be built…
🍿
English
@JerkPup @CryptoCyberia LLMs are notoriously bad at handling data. I wouldn’t be surprised if it hallucinated coordinates or some other shit like that
English
@CryptoCyberia Nope! You don't doubletap a military target. They're using vibewar as an EXCUSE.
It necessarily MUST be an excuse because its training data is previous airstrikes with human-generated weighting for which strikes are "good" and "bad". It doesn't generate policy it launders it.
English
Oh my fucking God, they actually did bomb that school because of vibe warring.
Jesus fucking Christ.
LLMs are being used for things like this, and they can't even beat Pokemon.
What the fuck.
Holly ⏸️ Elmore@ilex_ulmus
It’s time to quit, @AnthropicAI employees. You are in over your head.
English
@JustDeezGuy Now compare the latency of exchanging messages between Erlang processes and docker containers on K8s
English
This is why I’m unimpressed by Erlang/Elixir: every major language runtime has VERY high-quality M:N work-stealing “thread” schedulers with good APIs (structured concurrency), and the “isolated processes” and “RPC” got pushed up to an orchestration layer (DC/OS, Nomad, k8s…)
Anthony Accomazzo@accomazzo
Yes. There’s a reason you so rarely see the word “actor” in the erlang/elixir communities. The deeper, more general abstraction is the beam’s preemption and cooperative scheduler. Then you layer on processes with isolated memory. *Then* inter-process communication.
English
@paolino The Ruby/Steep version is an outlier because almost nobody uses it. I assume that's why the rest of the statically typed languages are much closer to dynamically typed languages
English
Types don't help AI, they actually hinder it.
Proof: look at Ruby with Steep - a Ruby type checker - and without! Or Python vs Python with mypy.
Token efficiency and expressiveness is way more important for AI generated code.
That's why Ruby was the best language in this test!
Yusuke Endoh@mametter
I benchmarked which language works best with Claude Code. Ruby, Python, and JavaScript came out cheapest, fastest, and most stable! See my article in detail: dev.to/mame/which-pro…
English
@e____f____e Are you paying enough for people to lose the 9:5 mindset?
English
that go into choosing the right environment. A lot of the times it doesn't matter too much, until it does and you hit a hard to overcome bottleneck
English
Which, of course, might not matter in some cases. If it's a web app you just fire gazillion instances and call it a day. But if it's not a web app? What if you can massively benefit from having access to shared memory instead of using a DB? There are lots of considerations 3/
English
I see people saying programming language choice doesn't matter anymore cause LLMs can code in any popular language. I think this is a bit weird. Even if we ignore the fact that LLMs perform differently in different languages, there are massive differences between some langs 1/
English
@glcst I mean, the difference is you usually want to have a webserver between Sqlite and the clients
English
Ok, so was the conclusion of this whole drama that SQLite is just fine and will perform just as well, with the added benefit of being able to support concurrent writes ?
Win for SQLite.
SpacetimeDB@spacetime_db
I am blind! I missed your blog post link. Actually your benchmark is strikingly similar to our methodology, complete with Pareto distribution! Several notable differences with our benchmark: 1. We put a web server/ORM in front of SQLite, since this is how SQLite is often deployed 2. We assumed many clients making one request at a time (no batching) 3. Write-only workload FWIW, we also benchmarked SQLite without a server and got around the 40k transactions you got here with no batching IIRC. So can SQLite hit SpacetimeDB-like numbers in this benchmark? Yes! But you have to be careful about how you go about it. What SpacetimeDB offers beyond SQLite is a webserver + static schema typing and client code gen + real-time subscription evaluation for arbitrary queries + distributed replication. That’s pretty much it. If you’ve got those things handled then SQLite is an excellent choice!
English