Lucas

senior swe © all tweets made high
se me levar a serio perdeu

lil thing -

[Lucas @lucasleandro]

lil gift
thanks @OpenAI

[Lucas @lucasleandro]

@appariciojunior @hschmaiske só uso o plano pro do codex, vale a pena

[Lucas @lucasleandro]

@hschmaiske @appariciojunior same

[Lucas @lucasleandro]

@TheRastrian @OpenAI eu investiguei legal antes de clicar no link

[Lucas @lucasleandro]

lil gift
thanks @OpenAI

[Lucas @lucasleandro]

@hackerdocc https://t.co/dV8pQwChJC

[jacky @jjacky]

whoever shared the "i have adhd" skill with me

thank you

it's made my claude replies so good

[Lucas @lucasleandro]

o mais complicado é se ligar no fato de que só da pra ter 1 (um) callback / app id

[Lucas @lucasleandro]

8 meses? kkkkkkkkk

[Micael Marques → WhatsApp para Devs @micaelmrsilva]

Conversei com uma empresa que estava a mais de 8 meses tentando integrar o WhatsApp direto na meta, e em 15 minutos conseguiram integrar usando a @ararahq. O pessoal deixa muita grana na mesa mesmo...

[Lucas @lucasleandro]

@hackerdocc I agree

[Lucas @lucasleandro]

@JohnPhamous please I don't want to merge

[Lucas @lucasleandro]

@VictorTaelin let me test it pf

[kitze 🇺🇸🦅🔫🍺🏈💸💪⚾️🏀 @thekitze]

im a fucking idiot for not getting a 128gb ram 2tb m4 max mac studio in january. absolute moron.

[Lucas @lucasleandro]

none, so i guess sysadmin

[Omkar @omkarships]

in 5 years, what will be more valuable?

writing code
reviewing ai-generated code

[Lucas @lucasleandro]

agora ta bom pra codar com AI, 5.6 foi o marco

[Lucas @lucasleandro]

i would like to see it doing the same work but 10x faster

[Mango @xmangonic]

GPT-5.6 is insanely good and efficient, but it still doesn’t feel like Mythos-tier.

For GPT-6, I don’t want another incremental upgrade. I want the most unhinged, closest-to-AGI model on the planet.

[Pranav Patel @ThatsPranav]

Local Development @nextjs vs @tan_stack A Side-by-Side Comparison

[Ryan Carniato @RyanCarniato]

Not sure who needs to hear this, but...

AI changes transport/distribution. It doesn't alter physics.

When code becomes cheap to produce, the limiting factor becomes how much complexity your architecture can absorb.

[Daniel San @dani_avila7]

If you recognize any of these names, you’re old school in the LLM world like me:

text-davinci-003, Bard, Claude Instant

[Lucas @lucasleandro]

acho que tem que ficar mais rápido

[Joao Vitor @joaovitor2763]

Tive minha primeira experiência onde gostaria que a IA tivesse feito cagada e ela não fez.

Dei uma task enorme pro Codex, quebrar vários arquivos de código para seguir o padrão e arquitetura definido.

Só que ao invés de fazer como deveria (um hotspot por vez, em vários PRs) dei um target + o guia de arquitetura e falei: faz tudo de uma vez só (vários commits, mas 1 único PR e sem parar).

Goal definido: 25 horas de trabalho depois, ~120K linhas de código , 335 commits, 600 arquivos alterados.

Aparentemente 0 erros. (ainda terminando de revisar, mas só de nada obvio ter quebrado, já estou muito surpreso).

Achei que ia ficar super feliz e animado com isso, confesso que fiquei meio "carai, que bad, where's the fun?"

Estranho, mas é isso.

[Lucas @lucasleandro]

I think it would be safe to assume that they want to keep the users in codex

[Howaboua @Howaboua]

People of Pi! If you're wondering why @pidotdev MIGHT be draining your Codex quota faster, here's your answer... Probably.

Unsure if this is 100% correct, but y'all remember when Gippity was running past 272k context and magically going to 115%? Was this on 5.4? My memory is hazy. 5.4 had 1mil context. 5.5 had normal 272k. It was always throwing at precisely 272k, never going above the specified max. Now that 5.6 has a 372k max, Pi lets it go over this. But OAI has admitted that going above 272k messes up the caching cost, as it's growing idk, exponentially? and they reverted temporarily back to 272. But Pi still isn't throwing, happily going to 120% and beyond, and only then auto compacting. Unsure if the limits are related or if there is another problem somewhere, or with just my setup.

This might be why your quotas are draining faster despite you having all these nice and long agent runs and not really doing anything differently.

Now this is a hypothesis, I cba to verify this. 373k has been merged back on GH Codex main, but it is NOT in prod, and is NOT what Codex is using rn.

"ok my dude, question. can you trace in the codebase, when since 5.6 got introduced has the default context limit changed? the history is: they launched 372. Ppl were complaining about super hard quota draining. They said they fcked up caching past 272k. So they reverted to 272k. You're saying they are back on 372 now. Is this reflected somewhere in the (Codex) codebase or what? :p"

(obvs some typos in numbers there, but you get the gist, typing too fast)

[Kahris @chrissotraidis]

I used Codex 5.6 Sol to get RollerCoaster Tycoon 2 running natively on iPad.

This is built on OpenRCT2, the open-source engine their team has poured years into. I compiled it to a native ARM64 iPad app and built the touch controls on top. No emulator, no x86, no streaming from a Mac.

Coaster building, scenarios, and park management all work by touch. First time I've seen RCT2 run natively on iPad.

Open sourcing it all below.

[Bilgin Ibryam @bibryam]

NEW: Durable Functions in PostgreSQL 🤯

pg_durable is a PostgreSQL extension. The workflow state, queue, retries, and recovery live in Postgres, and it uses Postgres durability / HA / backups / recovery.

https://t.co/ommJ4lCUdH