English
fried
697 posts
@nikitabier @justalexoki >Apps cannot access Apple’s FaceID API.
are you sure this statement is true?
English
optional face id to verify your twitter post before sending it is actually not a bad idea
Polymarket@Polymarket
JUST IN: Reddit CEO says the company is considering requiring Face ID to ensure humanity in order to crack down on AI bots.
English
@justalexoki @nikitabier you can skip it and use a passcode instead
English
@nikitabier that is absolutely available to apps. what do you mean?
English
English
I don’t actually want to crap on it as I get that it’s very good at what it does and saves software types a lot of time.
But it is an outside dependency and Wireguard isn’t that hard to set up. If you don’t know how, Claude/Codex can do it instantly.
I’ve worked from all over the world, just with my own Wireguard that took like 10 mins to set up.
But also, I’ve never come across a tool so universally loved by SWEs which is interesting.
Cynical take is that they don’t actually know much about networking but I am not a cynic.
But on the other hand I have never understood why I would want it, so maybe I’m confused lol.
English
@NickADobos @ankesh_anand Because they can use the larger pro model as a teacher model. Or even distill from Pro directly
English
@ankesh_anand This might be a dumb question, but why train pro first then?
Wouldn’t it make more sense to do those improvements on the small model first?
(Honestly maybe more of a random scheduling thing, but curious if this was planned or not)
English
"how can flash beat pro??" -> the answer is RL!
flash is not just a distilled pro. we've had lots of exciting research progress on agentic RL which made its way into flash but was too late for pro.
can't wait to finally bring them to pro👀
Lisan al Gaib@scaling01
Gemini 3 Flash scores higher than GPT-5.2, Opus 4.5 and Gemini 3 Pro on SWE-Bench Verified ???
English
@Qusaismael @T3chFalcon HTTPS is literally designed to mitigate a compromised gateway. Your traffic is encrypted from client to server. Nothing in between can decrypt it. Even more so if you use HSTS
English
@T3chFalcon But still, the traffic ( in a low chance ) in a public network can get a man in the middle in between, even if there is an HTTPS.
if you are on a compromised gateway.
English
@_robyn_smith try literally everything you have the opportunity to try.
English
@DennisonBertram @HanchungLee Bottlenecks are not an excuse to write slow code
English
@frieddriceeeeee @HanchungLee Memory management is probably the only one there that really feels like it makes sense to me. Codex might be fast, but inference is so slow so You don’t notice the benefits imho.
English
finally we have a company building cli using standard languages instead of brain rots.
go and python ftw.
github.com/MoonshotAI/kim…
English
@howmanysmaII @DennisonBertram @HanchungLee Python is definitely better than typescript for these CLI tools, although I never said I thought Python is appropriate either.
Codex does it right with Rust imo.
English
English
@DennisonBertram @HanchungLee slow, memory hungry, slop to read, infinite layers of abstraction and dependencies. The whole point of having a CLI tool is that it’s fast and lightweight.
English
@constantitus @vaxryy I can’t agree with this take. If you allocate memory you should free it. The OS memory management should be a last resort.
English
@vaxryy memory leaks don't matter in a tool that just runs for a few ms and then the os takes back all the memory. In fact, if it did actually worry about leaks and try to free them, then I would say it's poorly written.
English
at a single glance, there are like 7 direct memory leaks there. How is that a "well-written" tool?
The Lunduke Journal@LundukeJournal
Someone developed (and submitted) a “detect-fash” utility for systemd. “A utility to detect problematic software and configurations” It searches Linux systems for installations of the Ladybird Web Browser, Hyprland, or if the system is Omarchy Linux. It also contains code to detect if the user is @DHH (creator of Ruby on Rails & Omarchy). “systemd-detect-fash detects execution in a fascist environment. It identifies the fascist technology and can distinguish full machine fascism from installed fashware. systemd-detect-fash exits with a return value of 0 (success) if a fascism technology is detected, and non-zero (error) otherwise.” The code for “detect-fash” was ultimately rejected. However… Red Hat employee, and systemd developer, Zbigniew Jędrzejewski-Szmek reviewed the code and had this to say: “This is actually fairly well written and complete. Kudos! But such a tool is not a good fit for systemd.”
English
@SevDeutsch @teknium Reddit is definitely not the largest corpus of data for training LLMs. It’s significant but not the largest.
English
Not just yet- we have a benchmark coming out soon that leads frontier models to obtain 1% of the results that human software engineers get.
roon@tszzl
essentially all fixed time competitions at the edge of human skill have been grandmastered by machines, so labs must pivot to the only true challenge of unraveling the unsolved mysteries
English
@AravSrinivas Probably because you gave everyone a free year of Pro?
English
Actually, fastest growing app on both App Store and Play Store
Aravind Srinivas@AravSrinivas
Perplexity is the fastest growing GenAI app on Android 📈
English
@zephyr_z9 this is far too slow to make any meaningful difference
English
@garciamathias_ @ericzakariasson No. Cerberas just has custom inference hardware
English
@ericzakariasson aren’t large models artificially slowed down to make them feel more “intelligent”? models running on cerebras reply instantly, but that speed almost makes them seem less credible even though instant answers are what we actually want
English
Amazon so dumb like why not just take the money as soon as I make the purchase ?
English
I normally hate overengineering, but can't stop myself from doing it to my home network
English