sol t

Building autofictional hypernarrative @ | I use AI too much | previously 2x founder, stay at home dad, vc, big tech etc | not radical enough 🐜🐢

[sol t @solonteal]

More bullish on 4.5 than before tho

[sol t @solonteal]

GPT-4.5 → Conversational (use when I have a hypothesis)
GPT-4o → Creative (use when I’m figuring out my hypothesis)
GPT-o1 etc. → Computational (use when I’m delivering on my hypothesis)

[sol t @solonteal]

Biggest issue right now is 4.5's inference speed is so excruciatingly slow. I don't have even enough time to talk to the humans in my life, I don't want to be sticking around waiting for my AI daemon to form sentences no matter how vibe-y.

[sol t @solonteal]

Ok hard to place, but GPT4.5 might be so vibed up relative to 4o it's almost mechanical by comparison. the infinitely customizable machine, but you gotta give it super precise instructions.

wordcels rejoice, it's your time

[sol t @solonteal]

When GPT4o came out I went back to GPT4 for a bit. Feels similar with 4.5. Not in that it's worse, but the unstructured, sometimes erratic AI responses can sometimes have the best creative and inspirational nuggets for creativity.

[sol t @solonteal]

Better for my efficiency, fewer distractions, but I already wonder about the implicit increase of sycophantic relationships from 4.5. Making things look structured and flow like water makes it harder to figure out issues

[sol t @solonteal]

Before this scene: GPT 4o
After: GPT 4.5

[sol t @solonteal]

Oh, and o1Pro still definitely the smartest, but I find it better for professional writing vs vibes based freewriting

[sol t @solonteal]

30 min in I would describe the vibe difference with GPT4.5 as exchanging a lexus manual transmission car for a manual bmw.

Everything is magnified, need to figure out clutch sensitivity, matters if you want to look really good, i like…but still gas guzzler

[sol t @solonteal]

Another analogy: 3rd year of a foreign language (gpt4o) vs 3rd year + having lived abroad for a few months (gpt4.5).

Per always, realizing how much of my prompt eng/mind tuning has been to work around 4o's tendencies. Time to evolve my approach again.

Growing on me…

[Reworked @GetReworked]

Chain-of-Thought (CoT) Prompting Guide for Business Users by @solonteal https://t.co/7rNiJzj9sy #AI #PromptEngineering

[sol t @solonteal]

Not it. AGI delayed

[sol t @solonteal]

this could be it…

[VKTR.com @vktrnow]

What Are Philosophers Saying About AI? by @solonteal https://t.co/8hsZ5zLyc9

[VKTR.com @vktrnow]

Explore how Socrates might engage with AI chatbots, blending ancient wisdom with modern technology. VKTR Columnist @solonteal delves into practical applications for better responses and learning. https://t.co/cAa6Q1vrIK

#AI #AIInsights

[sol t @solonteal]

@Chris_Deutsch @CrainsChicago Free preview for those not in Chicago…https://t.co/shc4eXbEiz

[sol t @solonteal]

@wagner_tim Awesome news, right move, and matches the vision. Let’s go! 🚀

[Adam Mastroianni @a_m_mastroianni]

There are two kinds of problems: strong-link problems and weak-link problems.

Weak-link: quality depends on how good the *worst* things are

Strong-link: quality depends on how good the *best* things are