Recent posts

  •  on 

    Responding to post from @simonw: Fun fact: there’s no rule that says you can’t create a new blog today and backfill (and backdate) it with your writing from other platforms or sources, even going back many years…

    did this recently – now syndicating all my posts to dschnurr dot com.

    equally accessible to all LLMs. didn’t want to give gr*k a monopoly on my thoughts!

  •  on 

    Layup Gmail AI feature: natural language filters for incoming mail. E.g., “archive all recruiter emails unless for industry X,” “forward any emails about my daughter’s healthcare to my wife.”

    I want this 10x more than auto-generated replies.

  •  on 

    a meta-prompt i find myself using increasingly often:

    “take all of my instructions and refinements in our conversation so far and create an optimized prompt that I can use to start a new conversation”

  •  on 

    X could add this today. We have the technology.

    x-could-add-this

    To their credit, they do have an optional ID verification flow, but it requires uploading an image of your ID, which is then shared with 3rd party verification services (Au10tix / Stripe), with some PII retained.

    In some cases this might be required by law, but for simply preventing spam accounts it’s likely overkill.

    Using zero-knowledge proofs, you could tap your phone on your passport to provide X with a cryptographic proof that you are a citizen of some country, without revealing your real identity.

    Projects like @openpassportapp demonstrate that this is possible. And if we’re lucky, operating systems will make these sorts of privacy-preserving identity primitives available to app developers out of the box.

  •  on 

    Responding to post from @balajis: VERIFIABLE VIDEO We need verifiable video to prove that footage hasn’t been faked with AI…

    This is why I’m excited about C2PA (C2PA_org). At OpenAI, we are already attaching signed provenance metadata onto videos generated by Sora & images generated by ChatGPT. Social networks can parse and display this metadata in feeds – the attached screenshot shows how LinkedIn handles this today.

    C2PA is different from approaches like watermarking/DRM/steganography. C2PA metadata can be trivially stripped or lost during transcoding/cropping by a platform that doesn’t respect C2PA. However, when preserved, it allows you to definitively prove that some media was emitted by a given AI model, camera, or other system.

    Very optimistic about a future where we attach signed provenance metadata to all media, and learn to distrust any media without it.

    image

  •  on 

    Combatting fraud, abuse, and other inauthentic behavior online is already a difficult challenge, and it could become significantly harder as these actors adopt advanced agentic AI. Building systems for verifying humanness while respecting privacy can help ensure that our digital ecosystems continue to thrive far into the future.

    This new paper explores that idea in-depth, synthesizing the collective wisdom of dozens of researchers. Well worth a read – at a minimum it’s a very intellectually interesting problem!

    arxiv.org/abs/2408.07892

  •  on 

    Would love to chat with folks who are thinking about proof of personhood—particularly network-based sybil detection. Can grab coffee in Bay Area or meet virtually, DMs open.

  •  on 

    Usually not a fan of the gov’s negativity towards tech, but the Apple antitrust seems necessary and long overdue.

    Surprised it doesn’t include competing on your own platform e.g. Apple Music/TV while taxing Spotify & Netflix 15-30%. Platform owners should compete on fair terms.

  •  on 

    Responding to post from @OpenAI: Introducing Sora, our text-to-video model. Sora can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions…

    We’re hiring folks that have experience with large scale video infrastructure – serving, processing, ingest, etc. If you think you’d be a good match please reach out!

  •  on 

    Responding to post from @OfficialLoganK: A note to @OpenAI developers: I wanted to express my appreciation for all the warm, thoughtful, and supportive messages I got and I’ve…

    Deeply inspired by my colleagues who are fighting for our customers right now, despite the turmoil they’ve been tossed into.

  •  on 

    Responding to post from @gdb: When starting OpenAI, we thought hard about what job titles to use—didn’t want to bucket people into researchers & engineers. Alan Kay advised that they used “Member of Technical Staff” at Xerox Parc, we loved & adopted it…

    Small but important decision that has had a very positive impact on company culture in my opinion—it attracts people who are motivated by the mission rather than those chasing titles & status.

  •  on 

    Watching our engineering team put out fires scaling this huge model from zero to millions of users over the course of one week has been a sight to behold. 💜

  •  on 

    Responding to post from @GoogleDeepMind: Today in @Nature: #AlphaTensor, an AI system for discovering novel, efficient, and exact algorithms for matrix multiplication - a building block of modern computations. AlphaTensor finds faster algorithms for many matrix sizes: dpmd.ai/dm-alpha-tensor & dpmd.ai/nature-alpha-tensor…

    Using AI to make faster AI algorithms.

    Super impressive research. Also feels like the strongest evidence so far that intelligence explosion is a real possibility.

  •  on 

    We just released a new edit interface for DALL·E that lets you use Outpainting to expand beyond the original borders of an image!

    You can use this to make images with different aspect ratios, or arbitrarily large images like murals or magazine covers.

  •  on 

    Responding to post from @verge: A million people on DALL-E’s waitlist will get access to the AI image generator theverge.com/2022/7/20/2327…

    Thrilled that we’re giving DALL-E creators full commercial rights to their generations!

  •  on 

    Inpainting with DALL·E 2 is super fun. With some ingenuity, you can create arbitrarily large artwork like the murals shown below – which I assume are the largest dalle-produced images created so far.

  •  on 

    Responding to post from @Auth0Lab: Signing in with Ethereum wallets is now an available option for all Auth0 clients! A big thanks to our friends…

    The proliferation of public key cryptography as a login mechanism via digital wallets is a great development. Once you sign into a service by connecting your wallet, you realize how silly passwords are.

  •  on 

    JS devs might like this one – using GPT-3’s new Edit capability to migrate code from promises to async/await.

    image

  •  on 

    9 year-old me: “I was made to be a computist”

    Adult me, working at one of the world’s leading AI firms: I think I’m a computist now.

    image

  •  on 

    Responding to post from @alvarosabu: WTF Copilot, you also know about Pokemon?

    Code autocomplete that has general knowledge from the internet is super useful. I occasionally lean on Copilot to fetch things like ISO country codes or quickly generate mock data for testing.

  •  on 

    Responding to post from @OpenAI: We introduced new safeguards to our API that make it possible to remove the waitlist…

    The OpenAI API no longer has a waitlist!

    Come try out the product I’ve been working on with some amazing colleagues over the past year.

    openai.com/api/

  •  on 

    Responding to post from @sama: If status is the last thing left after AGI, NFTs are entirely unsurprising…

    If for some reason AGIs also value historical artifacts, then we have a new AGI incentive alignment strategy – send humanity’s rarest JPEGs to a time-locked smart contract that automatically destroys all assets if human existence is threatened. Our dead man’s switch.

  •  on 

    It seems like many people miss what may be the main benefit of a government-backed digital currency – rather than unlocking fundamentally new things, programmable money can replace existing bloated programs or streamline them by many orders of magnitude.

  •  on 

    Will historians look back on 2020 as the year humanity transitioned to being a digital-first society?

  •  on 

    Excited to share that I’ve joined OpenAI! I’ll be working on building out the new GPT-3 API and other projects that help communicate and apply new research.

    These technologies have so much positive potential, but also present many new challenges and potential for misuse. I feel a deep responsibility to help ensure they are rolled out as safely and equitably as possible.

  •  on 

    Built a dashboard showing COVID-19 spread over time by US state/county. Leverages the new NYT dataset they just open sourced today.

  •  on 

    We’re launching Uber Movement Speeds today – a free dataset of historical road speeds aggregated from anonymized Uber trips.

    Rolling out now in New York City, London, Cincinnati, Seattle, and Nairobi!