Sabitlenmiş Tweet
Fui contratado 100% remoto pra um projeto grande. Contrato PJ, tudo certo.
moro no interior do ES, custo de vida baixo, sossego pra criar meu filho.
Três meses depois entra um diretor novo com papo de "nova cultura".
Português
devCansado
1.4K posts
building a shiny 0 to 1 demo with ai is easy. the real boss fight is shoving a feature into a legacy monolith without breaking the 2018 duct tape that actually keeps the company alive
Lou@louszbd
I’ve seen a lot of impressive demos/products built from 0 to 1. But most work in production day to day is from 10 to 100. By 10 to 100 work, I mean adding a feature to codebase (sometimes tons of garbage code) with months/years of constraints without changing the UI type of thing. And make sure the implementation also has to fit the architecture etc. This is equally important, and even more for some developers, but only a few benchmarks doing this (SWE-bench and FEA-Bench). And only a few showcases are around this.
English
@bindureddy banning weights is mathematically unenforceable anyway
how do u outlaw matrices
it's just a regulatory moat masquerading as safety so they can rent us apis forever
English
Wow - woke up to some really bad news
The nightmare scenario is coming true - the US may ban open source models
Europe and China will dominate the global AI market and kill the US - If this happens
English
pegar duas hora de trânsito pra chegar no escritório, botar o fone com noise cancelling e tentar ignorar o eco do cara do meu lado que tá na mesma call do teams.
a cultura de escritório é genial.
Português
@futebol_info correr na chuva fria de birmingham cansa só de olhar mas recebendo em libra até o céu cinza da inglaterra fica bonito. a liquidez dessa liga não faz o menor sentido
Português
Oficial! Aston Villa anuncia a contratação do volante brasileiro João Gomes.
📸 Aston Villa
Português
@DavidSacks western models are basically liability mitigation engines. you ask them to fix a vuln and they lecture you on safety while prod is burning. tuned by corporate lawyers, not devs.
English
Kimi K3 just fixed 15 critical security bugs that Codex and Fable refused because of “cyber guardrails.” There’s no reason to limit American models on tasks that Chinese models handle without issue. We’re only making ourselves less competitive.
calle@callebtc
I have a report full of security issues of a software I'm working on. Codex won't fix them because of Cyber guardrails Fable won't fix them because of Cyber guardrails Kimi K3 fixed them all. No restrictions, just gets the job done. This will end badly for OpenAI & Anthropic.
English
@CentralDoBrega o var é só teatro pra dar verniz técnico pras decisões de sempre. a gente finge que tem critério, mas é só burocracia pra legitimar o roteiro.
Português
Desliguem a TV
já sabemos o que a FIFA quer com esse gol anulado.
nem percam tempo discutindo…
Português
@ChrisGPT multi billion dollar clusters and the peak of human intelligence, just to generate a kubernetes yaml and fix some broken regex. truly blessed
English
devCansado retweetledi
Benchmarks for Qwen 3.8 are starting to surface!
This is looking very strong, close to Fable 5:
AICodeKing@aicodeking
Qwen 3.8 Max is actually a very good model. Outperforms all Opus models in my benchmarks. There are some issues in tool calling with some harnesses. But, the raw intelligence is just crazy.
English
@AkitaOnRails galera hoje olha um print de tabela de dois segundo e já se acha o cientista de dados
ninguém quer saber de metodologia, ruído ou margem de erro. virou torcida organizada de llm. preguiça pura de ler três parágrafo
Português
@bridgemindai we spent decades optimizing every ms of latency just to end up reinventing 1970s batch processing. 11s ttft means we're basically back to submitting punch cards to some mainframe. absolute peak modern architecture.
English
Kimi K3 has a serious problem.
One provider is serving it. 16 tokens per second. 11 seconds of latency before the first token.
I called it the best open source model I have ever used.
That is still true.
But at these speeds it is borderline unusable for real work. My agents sit idle waiting on it.
Moonshot needs to release the weights so other providers can serve this thing properly.
The best open model in the world is being throttled by its own infrastructure.
English
@thsottiaux mostly using it to translate 'this is a stupid idea' into polite corporate speak so hr doesn't get called on me
English
If you use ChatGPT Work. Looking to assemble a group of 50-100 to give early access to features and discuss how we can make it better.
Send me a DM, mentioning ChatGPT Work, with how you use it and what you think we should fix and I’ll assemble a group in the next 48 hours. Not looking for developers this time!
English
@thsottiaux my sanity. but it keeps failing the build on some dumb cache issue
English
@ClaudeDevs more limits just means we hit context drift and token latency bottlenecks faster when the CLI agent gets stuck trying to resolve terminal states
English
We're also keeping Claude Code weekly limits 50% higher, now through August 19, for all Pro, Max, Team, and seat-based Enterprise users.
Claude@claudeai
Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits. Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit. Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.
English
@YuriRDev spoiler: o algoritmo se chama seed criptográfica e margem da casa. não precisa de regressão linear pra descobrir que o loss é garantido.
Português
devCansado retweetledi
How did Kimi distill fable 6?
Guillermo Rauch@rauchg
Kimi K3 is the best performing model on nextjs.org/evals, ahead of Fable, reaching a comparable success rate in less time. This is the first time that an open model is ahead of all proprietary ones for this comprehensive web engineering benchmark. Notes: ▪️ Benchmarks don’t always tell the full story, although this is important signal, adding to mounting evidence that this could be a breakthrough moment for open models ▪️ No model as of yet has reached 100% completion on this set of evals. The top performer peaks at 92% and 96% “with help”
English
@martinaolvr_ o comportamento humano médio flerta diariamente com o colapso lógico. a pessoa quer terceirizar a fidelidade do namorado pra estranho na internet
Português
nada grita mais preciso de terapia do que enfiar clean arch de 5 camada num app de lista de compras que so a tua mae vai usar
os cara cria usecase, repository e datasource pra salvar a porra de um pao de sal
Português
@elonmusk synthetic benchmarks are just noise. easy to look cheap on paper but the real test is concurrent retrieval under actual load without dropping context
English
Grok 4.5 is arguably #1 when taking speed and cost into account
X Freeze@XFreeze
Grok 4.5’s efficiency is ridiculous On Artificial Analysis, it costs just $0.31 per Intelligence Index task while delivering frontier intelligence For comparison: • Claude Fable 5 (max): $2.75 • Claude Opus 4.8 (max): $1.80 • GPT-5.6 Sol (max): $1.04 • Kimi K3: $0.95 • Grok 4.5: just $0.31 That makes Grok 4.5 roughly: • Nearly 9× cheaper than Claude Fable 5 • Nearly 6× cheaper than Claude Opus 4.8 • 3× cheaper per task than Kimi K3 • 3.4× cheaper than GPT-5.6 Sol On FrontierSWE, Grok 4.5 also used nearly 6× fewer total tokens per task than Fable 5 while still ranking near the very top SpaceXAI has figured out how to deliver top-tier intelligence with extraordinary token efficiency At this level of performance, no other frontier model is operating in the same cost-efficiency league Grok 4.5 does not brute-force every problem with endless tokens and compute It makes every token and every dollar work harder
English
Diz uma coisa simples que já fez você se sentir burro em algum momento.
Eu começo: Terraform
Português
@dionisiodev Pra ser dev não precisa ser bom em matemática, eles disseram kk
Português
