Sup*

Casted a spell so only cool people see this profile. Did it work?
I post about AI, evals, hackathon projects, vibe coding and things that make me laugh.

[Sup* @mostlyaboutai]

https://t.co/fYopxOTbx6

[Sup* @mostlyaboutai]

@contraben It’s because we look at AI the wrong way.
Novel prompts = novel outputs.

Judging models based on “Generate an award-level landing page for XYZ” level prompts is like asking the genie “give me more money” and then complaining you didn’t get enough.

UI Writing is a skill.

[Sup* @mostlyaboutai]

Are you cuhl enough!?

[Jitter @jittervideo]

New this week: Jitter MCP (beta) 🔌

Your favorite AI tool, now with Jitter.

Live now👇

[Sup* @mostlyaboutai]

A question for the companies that are killing AI Slop:
How do you even decide what's slop and what is not?

[sunny madra @sundeep]

Goodbye traffic jams 🤣🤣🤣

[Sup* @mostlyaboutai]

@krea_ai gives the perfect skin................

[Sup* @mostlyaboutai]

@erinnhorner @contra @contralabs It's only been a short time since I discovered Contra, but the community and the sense of belonging the team creates for its users are what kept me hooked. And I think that's one of the biggest successes of your work as VP of Brand.

Wishing you the very best, Erin!

[Sup* @mostlyaboutai]

@MagicPathAI is 'that' tool!

[Pietro Schirano @skirano]

Since we launched MagicPath in ChatGPT, we have been adding 1 million designs a week to the platform. For context, it took us one year to get to 3 million.

[Sup* @mostlyaboutai]

I hate the direction AI is heading in.

It feels like a constant fight for incremental progress. Forget the benchmarks for a second. When was the last time something genuinely exciting came out?

Agents were the last real innovation that made me think, "this changes everything"

[Sup* @mostlyaboutai]

People giving the same prompt to two different AI models and jumping to coclusions are **DUMB**

[Sup* @mostlyaboutai]

@notjazii It is EXPENSIVE though! Especially on api

[Sup* @mostlyaboutai]

WOW
LLMs trained on simulated faulty hardware can get more resilient to bit flips as they scale. They effectively learn error-correcting codes whose overhead stays finite.

If this holds, we could run inference on far cheaper, and on lower-energy without the usual accuracy crash.

[Sup* @mostlyaboutai]

I’m so confused and terrible at remembering names lately that I’ve been calling Google’s Omni as “Omma” which belongs to spline.

[Sup* @mostlyaboutai]

@theo It is great indeed but not S tier.

[Sup* @mostlyaboutai]

@vamsibatchuk Trust

[Sup* @mostlyaboutai]

@sundeep Agreed, this is a better wtg for them as well as the industry as a whole

[Sup* @mostlyaboutai]

@thsottiaux @OfficialLoganK Salty

[Sup* @mostlyaboutai]

@gregisenberg Overall there is so little 'innovation' happening rn, that innovative products now hold more value than ever before.

[Sup* @mostlyaboutai]

@gregisenberg The funny thing is that almost all these advancements are happening within the same industry, with competitors constantly launching new features just to match/outmatch each other.

[Sup* @mostlyaboutai]

I wanna try Muse wonder rn 😶‍🌫️