Winston Howes

Build long-running agents with more control over agent execution. New capabilities in the Agents SDK: • Run agents in controlled sandboxes • Inspect and customize the open-source harness • Control when memories are created and where they’re stored

So excited to share that we're bringing Computer Use to Codex. Computer Use lets Codex see, click, and type into your Mac apps, with its own cursor. It's a magical feeling to have agents using your apps in the background, and still get to use your computer at the same time.

.@OpenAI blog post: Designing Agents to Resist Prompt Injection We’ve found that prompt injections are evolving to look more like social engineering over time as models get smarter and mitigations need to evolve too. openai.com/index/designin…

🚀 health sync is out Analyze your health data locally across providers. It syncs data from Oura, Withings, Hevy, Strava, and Eight Sleep (more coming) into a local database you fully control. I run it on a cron and use @openclaw to generate daily summaries of my activity and sleep. github.com/filipe-m-almei…

We're taking ChatGPT privacy to the next level. OpenAI is putting together a founding team to bring advanced encryption-based privacy to ChatGPT, the OpenAI API, and our future consumer devices. If you are an expert in TEEs and E2EE, or a product engineer who cares deeply about privacy, reach out!

@autosourcer0 @GladiaLab "almost surely injective"

@winhowes @GladiaLab small enough hidden state and it would again be untrue

LLMs are injective and invertible. In our new paper, we show that different prompts always map to different embeddings, and this property can be used to recover input tokens from individual embeddings in latent space. (1/6)

@autosourcer0 @GladiaLab injective on hidden state, not outputs

@GladiaLab i dont see how you can prove they are injective without keeping the weights fixed and testing all the permuted inputs say the llm is trained to only output between two response tokens as a choice - clearly, the prompts are not injective

@Endothermia @JustinAngel @GladiaLab A collision just proves that it's not injective, but it doesn't prove that the average case isn't invertible

@JustinAngel @GladiaLab prompt -> response is obviously not invertible. It's trivial to construct two prompts that yield the same response.

@chaddepue But my point is more broadly the government's actions here are hypocritical to how they talk about Charlie

@chaddepue He branded himself (and is widely seen on the left and right) as pro free speech, but the lists (while I get the reasoning) had negative second order effects; good they basically died out. Government pressure on speech though is very different: direct suppression.

@chaddepue @enortham @filipe_almeida @antonellasturn Congratulations you two and you're welcome. 🫶🏼

@johnpalmer Gpt5pro

Who’s the best website / marketing site developer for hire? Not designer. Developer,

GPT-5 rollout updates: *We are going to double GPT-5 rate limits for ChatGPT Plus users as we finish rollout. *We will let Plus users choose to continue to use 4o. We will watch usage as we think about how long to offer legacy models for. *GPT-5 will seem smarter starting today. Yesterday, the autoswitcher broke and was out of commission for a chunk of the day, and the result was GPT-5 seemed way dumber. Also, we are making some interventions to how the decision boundary works that should help you get the right model more often. *We will make it more transparent about which model is answering a given query. *We will change the UI to make it easier to manually trigger thinking. *Rolling out to everyone is taking a bit longer. It’s a massive change at big scale. For example, our API traffic has about doubled over the past 24 hours… We will continue to work to get things stable and will keep listening to feedback. As we mentioned, we expected some bumpiness as we roll out so many things at once. But it was a little more bumpy than we hoped for!

@xwang_lk Version numbers

Really? Still?

Looks like everyone's excited across the industry this week. Good good

@LittleJoeTables @IAmMandatory @bishopfox @OpenAI Welcome aboard!

After 14 years, today was my last day at @bishopfox, and I'll be joining the OffSec team at @OpenAI next week!

@lcamtuf Raccoons

Kudos to @Snapchat and specifically @GoodLuckAmerica for seeming to have the most balanced reporting on the LA protests so far. Not sensationalized just facts

@lcamtuf I clicked the post