Sean

[Sean @realSean_Murphy]

No matter your perspective on AI, nobody benefits from misaligned systems. As AI grows more capable, the goal is to maximize its upsides while actively mitigating its harms. We can push innovation forward, provided we institute clear, enforceable safety benchmarks every step of the way.

[Sean @realSean_Murphy]

@CultureCave_ If the books ever get released* (which isn't likely)

[Sean @realSean_Murphy]

@gao147357 @marishad5gh They gotta get the money shot for their publishers.

[Sean @realSean_Murphy]

@FFunkdom @marishad5gh Seems like both.

[Sean @realSean_Murphy]

@lower_votingage @WesRoth I was in denial. I didn't expect it to be true. The rate of progress is astonshing and the benchmark has only been out for roughly ~6 months.

[Sean @realSean_Murphy]

@lower_votingage @WesRoth No, he's making the numbers up.

[Sean @realSean_Murphy]

I built an AI-powered typing software designed for a 4-input sip-and-puff device. By using a hard puff, soft puff, hard sip, and soft sip, it maps directly to a custom alphabet layout and uses an LLM to instantly decode your text.

Here is exactly how it works: 👇

���️ The LayoutThe alphabet is divided into three main rows, each assigned a number (with the fourth input acting as a zero/space). In my setup, it looks like this:
Row 1: c, f, h, i, j, k, l, n, u, w, x, y

Row 2: a, b, d, g, q, r, t, z

Row 3: e, m, o, p, s, v

🔢 The InputWhen typing, you aren't selecting the exact letter. You are just using the sip-and-puff device to select the number of the row that holds your letter.

So, if you want to type a sentence like, "I need water, my throat hurts," you would input this exact sequence: 1013320122320310212322011223
🧠 The AI DecodingOn its own, that string of numbers is completely ambiguous. To solve this, I wrote a Python script that takes your number sequence and feeds it into an LLM through a custom prompt. The model uses predictive context to instantly decode the sequence and output the exact sentence you meant to type.

I have it fully set up as a Python script right now. If you want to use it or test it out, reply or shoot me a DM and I can provide the code and walk you through how to run it! 💻
#Accessibility #AssistiveTech #PromptEngineering #Python #DisabilityTech #TechForGood

[Sean @realSean_Murphy]

My weird obession with developing a scifi mask that you can communicate with by blowing into it led me to developing a way you can write out entire sentences with just 4 numbers.

[NVIDIA AI @NVIDIAAI]

Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark.

NVIDIA AVO completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals.

[Sean @realSean_Murphy]

@BioLLM_ @BlackrockNeuro_ I can show you how to do it without an implant.

[Sean @realSean_Murphy]

Prompt engineering proved useful for me. I was able to figure out a way to use the number pad on a TV remote to type out words with the help of Gemini, so I didn't have to use that on-screen keyboard (which I find slow and a pain in the ass). You can see in the screenshot below that it can decode the sequence back into words.

[Sean @realSean_Murphy]

@MarvelStudios I can show you guys how to use the number pad on your TV remote so you can search for the trailer without needing to use an on-screen keyboard (which is a pain in the ass) by utilizing Gemini.

[Sean @realSean_Murphy]

@ChiseledAdonis Bro might want to put that down before he gets hit by 5 gallons of regret.

[Sean @realSean_Murphy]

@Rainmaker1973 If you can't even fill in four words you must be retarded lmao.

[Sean @realSean_Murphy]

@AJAlkaline @KobeissiLetter That's why we need models on-device.

[Sean @realSean_Murphy]

@OfficialLoganK @an_engineer_log When will we get a model like Nano Banana Pro on device?

[Sean @realSean_Murphy]

@emollick The image generation feature in both Flash and Pro has deteriorated to the point of being practically useless. I can't get the model to generate anything consistently. When I ask it to adjust the hair or turn the face slightly, it generates an entirely different person.

[Sean @realSean_Murphy]

@NoahKingJr ANI -> AGI -> ASI -> ASDI?

[Sean @realSean_Murphy]

@4R7I5T And I have your plan B

[Sean @realSean_Murphy]

@Crystal74063093 @Rainmaker1973 @grok Stop relying on Grok to examine these videos, and just use your eyes, fool. Models tend to hallucinate, and we don't have the same verification methods for video that we do for text.