Augustus (gsd)

algorithms given agency must be grounded by ethical ideas.

opinions and ideas are anonymously my own — not an agent’s ⚖️⚙️

CS PhD student

[Augustus (gsd) @algoretic]

https://t.co/s1b7DSFWrc

[Augustus (gsd) @algoretic]

@OscarMDavies AI writing repeatable, auditable algorithms to do things is almost always preferable to agents acting on their own - they are non-deterministic.

[Augustus (gsd) @algoretic]

@KyleTibbitts The nuclear power plants have meaningful physical constraints and a thorough regulatory framework built around them. AI does not, and perhaps can not, have either of these.

[Augustus (gsd) @algoretic]

@elonmusk It's AI generated. lol

[Augustus (gsd) @algoretic]

@quietforgelab Let's connect 🤝

[Augustus (gsd) @algoretic]

@ElonSilencedMe_ @ewarren In the hypothetical universe where they pass popular legislation before the next election.

[Augustus (gsd) @algoretic]

@juliantang Hi Julian, let's connect. 🤝

[Augustus (gsd) @algoretic]

@Mitchell_Suen Hi Mitchell, let's connect. 🤝

[Augustus (gsd) @algoretic]

@1k_clouds Hi Cloudius, let's connect. 🤝

[Augustus (gsd) @algoretic]

@hrshsn_ Hi harsh, let's connect.

[Augustus (gsd) @algoretic]

@SardiOps Hi Sardi, let's connect.

[Augustus (gsd) @algoretic]

@Johnlin1985 Hi John, let's connect.

[Augustus (gsd) @algoretic]

@fabian_bao Hi Fabian, let's connect.

[Augustus (gsd) @algoretic]

@TimofeyBara Hi Timo, let's connect.

[Augustus (gsd) @algoretic]

@Yue1650749 It's also fun to dig into code and find that researchers selected seeds 5, 6, 137, 534, and 1011 for... totally not picking seeds to make our results better, we swear.

[Augustus (gsd) @algoretic]

@daniel_p_gross Something that we need to be careful about with this kind of analysis is that people write abstracts in ways that are in line with the way AI generates text. I have had entirely human-written abstracts identified at >65% AI-generated. Pangram is limited in its specificity.

[Augustus (gsd) @algoretic]

@SandroAmbuehl If I see slop, I skip slop. that simple. don't have slop in your abstract!

[Augustus (gsd) @algoretic]

@SamVale29 Strongly agree. It is very easy to configure the Claude app to do many things that apps are selling: e.g. my AI trainer.

[Augustus (gsd) @algoretic]

@paulnovosad The bottom line is absolutely right, and there are several research studies that back him up.

[Augustus (gsd) @algoretic]

@justinskycak For the vibe coders, it is a rude awakening after a few days, when the repository exceeds agents' context limits.

[Augustus (gsd) @algoretic]

@pabloolpeb @OscarMDavies Absolutely. A* search and the perceptron are also AI, of course, and despite their simplicity have likely led to developments in all three areas the author names.