Joseph shunk

Software Engineer

[Joseph shunk @Shoejunk]

@MelvinInvests Jensens letter is NOT an endorsement of Moonshot. "...unlawful efforts to extract value from closed models raise legitimate concerns. Those concerns should be addressed through targeted legal and commercial frameworks..."

[Joseph shunk @Shoejunk]

@iruletheworldmo GPT 6, Fable 5.1, DeepSeek, Gemini 3.5 Pro

[Joseph shunk @Shoejunk]

@Dimillian Oh! Didn't realize you can use voice remotely. But it's using a different...voice. It's not respecting the voice I picked either on my phone or on my desktop.

[Joseph shunk @Shoejunk]

@SuryaRajendhran @camsoft2000 @OpenAI Yeah, I don't know if it's possible. I was curious about that myself.

[Joseph shunk @Shoejunk]

@SuryaRajendhran @camsoft2000 @OpenAI You have to start a new chat ("New chat"). Then you can start a voice chat from there. The voice chat isn't tied to a specific project. It works across all of Codex and will send tasks to individual projects as needed.

[Joseph shunk @Shoejunk]

@SebAaltonen Pirating books should be illegal. But I don’t think the market is sustainable if distilling is allowed. Moonshot distilled Fable at a fraction of the cost and gave it away. What is the incentive for companies to pay upfront training cost if it will just get immediately distilled?

[Grim Dawn @GrimDawn]

The wait is over! Grim Dawn - Fangs of Asterkarn is out NOW on Steam & GOG!

Thank you for taking this ride with us for the last decade! Now venture into a harsh new realm, master the Berserker, conquer Ascendant Mode and do so much more!

Your next chapter in Cairn begins today!

[Joseph shunk @Shoejunk]

@iruletheworldmo @PeterDiamandis However long it takes to distill a model.

[Joseph shunk @Shoejunk]

@chamath You're talking about decreased model layer cost but it's only cheaper for Moonshot because they are stealing the work of Anthropic. Someone still has to pay the cost of the original train and who would do that if distilling is allowed?

[Joseph shunk @Shoejunk]

@johnennis @filicroval @kimmonismus True and Moonshot can also train off communal knowledge, but once you train a model, you added value; the result is no longer communal. It's like growing fruit using the sun and rain. Sun and rain are communal but producing fruit still takes work; I can't just take your fruit.

[Joseph shunk @Shoejunk]

@filicroval @kimmonismus The limit of what Claude can be used for is up to Anthropic's terms of service. They're allowed to decide how they want their model to be used. Most of the data used to train Claude was public data that was legal to use. Some was not legal and they paid fines for it.

[Joseph shunk @Shoejunk]

@niccruzpatane @Tesla Sounds kind of like you’re saying FSD doesn’t slow down or avoid potholes or rough roads which is why it works well with the Cybertruck because you can’t feel the bumps.

[Joseph shunk @Shoejunk]

@matt503ea5sf9z5 By copying Claude

[Joseph shunk @Shoejunk]

@thsottiaux @jxnlco Live voice model interacting with work and codex

[Joseph shunk @Shoejunk]

@bridgemindai This is dangerous. It feels unlimited until you run out halfway through the week. Better save up your resets and spread out your usage carefully.

[Joseph shunk @Shoejunk]

@DanielMiessler Or they could release Opus 5

[Joseph shunk @Shoejunk]

@rezoundous I want to use Sol High but it burns through my limits too fast. Trying out either Sol Medium or Luna Extra High, and moving up to High if those don't do the job.

[Joseph shunk @Shoejunk]

@elvissun It's between sol medium and sol high according to DeepSWE.

[Joseph shunk @Shoejunk]

@huskirl @modnareman0 I can confirm. The new voice model gets this wrong on the highest intelligence level for me as well. What I think is happening is that the voice model only hands off tasks to GPT 5.5 if it recognizes that the question is difficult, and in this case it didn’t think it was.

[Joseph shunk @Shoejunk]

@OpenAI When is it coming to Codex?