🌋Victor Hugo Germano

Caitlin KALINOWSKI over X I resigned from OpenAl. I care deeply about the Robotics team and the work we built together. This wasn't an easy call. Al has an important role in national security. But surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got. This was about principle, not people. I have deep respect for Sam and the team, and I'm proud of what we built together.

OpenAI head of robotics just resigned over company deal with the Pentagon saying…

“Surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got”

Adversarial Attacks with Carlini & Wagner Approach Adversarial Attacks with Carlini & Wagner Approach In the ever-evolving landscape of artificial intelligence and machine learning, the quest for robust and reliable models is relentless. However …

quase - entrego no fim do mes de abril...

Mas pesquisa parte do trabalho e da implementação Carlini-Wagner, se quiser referências:
medium.com/@zachariahar...

imagem de resultado do notebook, que usa um audio que é interpretado pelo whisper e transcreve uma frase específica...

Você preocupado com IA te substituir, e com os agentes tipo openClaw.

Meu projeto no mestrado: prompt injection em mensagens de audio, que pode impactar qualquer agente com voice assist que use o whisper, da openai...

People Are Calling Meta Ray-Bans "Pervert Glasses" On Bluesky, users quickly embraced the term "pervert glasses" to refer to Meta's Ray Ban smart glasses, following a shocking investigation.

Make it stick.

Type of guy who cosplays his software having feelings in order to avoid feeling personally responsible for how it was used to murder children is an increasingly common kind of guy in the software industry

Screenshot of headline: Microsoft gets tired of Microslop, bands the word on its Discord, then locks the server after backlash

Lmfao this is so fucking sad 😂😂😂

Buddy Guy: Tiny Desk Concert YouTube video by NPR Music

Did you know that "young Sammie" and "older Sammie" did a Tiny Desk today?

Yep, Miles Caton and Buddy Guy performed together

#Sinners

youtu.be/m5XxOLdMSS8?...

Debugging feels like trying to find a needle in the haystack. Only, the needle is invisible, the haystack is on fire, and I'm not sure the needle was there in the first place.

Exclusive | OpenAI Employees Raised Alarms About Canada Shooting Suspect Months Ago The ChatGPT maker opted against informing Canadian authorities about Jesse Van Rootselaar’s descriptions of violence last June.

Absolutely wild. WSJ reporting that some OpenAI staffers thought the Tumbler Ridge shooter's use of ChatGPT raised the potential of real world violence but leadership decided not to alert RCMP. www.wsj.com/us-news/law/...

news: today capitalist AI did something fucking terrible, just like yesterday
some lefties: what the fuck, this capitalist bullshit sucks
AI pushers: can't believe the left hates technology

But are the two companies really so different at the end of the day? Ed Zitron, a prominent AI commentator and tech PR specialist, argues there’s little daylight between them — and that the disagreements are trifling, at best. “They are very similar companies doing very similar things with very similar goals, which is to enrich their executives and get as much media attention as possible,” Zitron told DFD. He pointed out that Anthropic and OpenAI have both lobbied for export controls on semiconductors, and similarly pushed the idea that the U.S. needs to boost the AI sector to compete with China. They advocate for safety rules at various times, some of which have caught on in state legislatures, though Zitron contends that there’s little possibility that those efforts will gain much traction on the federal level given the hesitance around tech regulation in the current administration. Zitron said, “Dario Amodei and Sam Altman are functionally the same guy — they just look a bit different.”

Talked to Politico about Sam Altman and Dario Amodei being the same guy
www.politico.com/newsletters/...

Não precisamos aceitar a “IA” como inevitável na educação Elixir and Erlang enthusiast. Associate Professor at UTFPR

A oficina 'Não precisamos aceitar a “IA” como inevitável na educação' está chegando...

Coloquei um link para um PDF em construção

adolfoneto.elixiremfoco.com/extension/co...

India’s ‘AI Impact Summit’ Promises Little More Than Spectacle The fourth in a series of convenings that kicked off in Bletchley Park in 2023, the summit is the first held in the Global South, writes Apar Gupta.

"Any claims of AI solving for population scale problems must be matched against the reality of the Global South, in which young boys were applying enamel paint on the railings... with their bare hands to prepare for the arrival of Silicon Valley CEOs." www.techpolicy.press/indias-ai-im...

Foto de uma paisagem de montanha com uma grande pedra ao fundo, e um grande jardim com floresta. Um cavalo pasta ao longe

Tem coisa que o Brasil dá um baile!

Conto de fadas com torresmo, café e queijo meia cura 🥰🇧🇷

#photography

"due to fundamental mathematical constraints that cannot be solved through better engineering" <- a lot of linguistic gymnastics to say that "hallucinations" are a feature, not a bug

Quase carnaval…

The primary use case for this product is to erode consent.

a man with a beard is making a funny face while wearing ear buds . ALT: a man with a beard is making a funny face while wearing ear buds .

#fantastico faz uma MEGA desserviço à discussão sobre inteligência artificial, multiplica o hype e deixa de discutir claramente o que houve com o moltbook… que bosta

pra variar, no fim são humanos nos bastidores...

Hater Season: Openclaw with David Gerard Podcast Episode · Better Offline · 02/04/2026 · 38m

Here’s this week’s Better Offline, I’m joined by @davidgerard.co.uk for Hater Season to talk about the stupidity of Openclaw, how seemingly smart people keep being “one-shotted” by AI, and why we’ve been headed toward a calamity in the markets since 2008.
podcasts.apple.com/us/podcast/b...

crypto wouldn't be falling like this if trump was president

É uma abordagem simples mas super interessante! Estratégia de manutenção de contexto e organização de tasks para que as instâncias consigam rodar “quase” sem problema.
Isso não saiu com vibe coding, mas foi bem facilitado com ferramentas de auto complete

GitHub - victorhg/ralph: Implementation of the Ralph loop for agentic Implementation of the Ralph loop for agentic. Contribute to victorhg/ralph development by creating an account on GitHub.

Criei uma versão do “Ralph Loop” que roda localmente usando uma instância Ollama.

Uma abordagem mega interessante pra executar tarefas autonomamente… o tal #agentic flow

Usa bash e python…

github.com/victorhg/ralph

DHS: Want affordable housing? Help report illegal aliens in your area. Call 866-DHS-2-ICE. If you are an illegal alien, you can take control of your departure using DHS.GOV/CBPHOME.

straight up Gestapo shit

Janeiro, geralmente ligado a férias e descanso, nos últimos anos têm registrado mais e mais episódios de ataques e tentativas de rupturas institucionais.

Coluna de @fabi2moraes.bsky.social 🧵

ICE Is Using Palantir’s AI Tools to Sort Through Tips ICE has been using an AI-powered Palantir system to summarize tips sent to its tipline since last spring, according to a newly released Homeland Security document.

NEW with @makenakelly.bsky.social, ICE is using generative AI from Palantir to sort through immigration enforcement tips and summarize them. Details here:
www.wired.com/story/ice-is...

whoever came up with “create the problem, sell the solution” probably didn’t expect someone would figure out how to sell the problem too

"using LLMs, as well as being funded by or working for AI companies, are conflicts of interest for any academic, which must be declared with every published piece of work"

Sorry, Jensen, can't make us stop. Mystery AI Hype Theater 3000 will continue on with the #RidiculeAsPraxis

cc @alexhanna.bsky.social

It's important to continuously remind people it is impossible for guardrails to prevent this. LLMs can't have perfect guardrails. The idea is they can meaningfully reduce risk, however the existence of an entire subfield about jailbreaking and prompt injection amplifies that same risk.