William Sousa

growth x creative • ai • web3 @noncitizen_

[William Sousa @willw3b]

Midjourney went from from rendering things that don't exist to revealing things that do.

[Midjourney @midjourney]

Announcing a new division of Midjourney called "Midjourney Medical"

[William Sousa @willw3b]

Moved USDC from MetaMask to @cryptocom and @okx.

On-chain transfers were instant.

CEX side:

- 25 min delay
- sender full name required
- home address required
- sender wallet type required

We built tech that settles globally in seconds, then wrapped it in compliance friction that can feel worse than legacy banking.

I’m not anti-regulation. But CEXes need to start pushing back on what this is doing to the UX.

[William Sousa @willw3b]

Rio claimed they built a groundbreaking frontier 397B model.

Nex checked the weights and found it was mostly their N2 Pro + Qwen 3.5.

The model even started calling itself Nex without a prompt. 💀

They updated the card after getting exposed.

Remixing is fine but give credit where it's due.

[Nex @NexEcosystem]

The Rio 3.5 model broke the internet this week. The plot twist? It’s essentially our open-source model, Nex N2 Pro, wearing a different hat.

🤯 We analyzed the weights, and the recipe is exact: Rio 3.5 ≈ 0.6 * Nex N2 Pro + 0.4 * Qwen 3.5

It even literally introduces itself as "Nex N2 Pro" if you ask it without initial system prompt!

😂 We are flattered that the City of Rio used our work to achieve SOTA performance. Thanks for the ultimate benchmark validation.

🤝 But in the open-source world, attribution matters.
👇 Full mathematical proof & verify script in the first reply!

[William Sousa @willw3b]

@bidcapbro @HermesAgentTips Yes I’m aware.

[William Sousa @willw3b]

Cheaper tokens do not always mean a cheaper workflow.

I’ve been noticing this using Hermes and open-weight Chinese models.

Sometimes the cheaper model needs longer prompts, more retries, more tool calls, or another model to fix the output.

So the token price looks better, but the workflow is not always cheaper.

That part gets lost in a lot of benchmark charts.

[Hassan @buildwithhassan]

opencode published their real model usage data. what developers actually run when they're paying for it:

1. deepseek v4 flash: 32T tokens
2. deepseek v4 pro: 19T tokens
3. kimi k2.6: 6.5T tokens

deepseek is running more tokens than the next 16 models combined. it's actual usage from developers spending their own money.

glm-5.1 grew 419% too. the models winning on price and reliability aren't always the ones winning on twitter.

[William Sousa @willw3b]

The same vibe-coder engineers using Opus for everything would hire a private jet to cross the street if someone else was paying.

Kimi 2.6 and DeepSeek v4 Pro can handle a lot of the grunt work.

Save the expensive models for the parts where being slightly wrong costs more than the tokens.

[Polymarket @Polymarket]

NEW: Uber is reportedly capping employee use of AI vibe-coding tools at $1,500 per month after blowing through its AI budget.

[William Sousa @willw3b]

@VictorInFocus @krea_ai this is great. nice move on having the kids wear mask (I take it it was an artistic decision to avoid the lip sync AI issue?) you also nailed to the realism 👌

[William Sousa @willw3b]

you gotta watch this.

[PJ Ace @PJaccetturo]

This is one of the best short films I've seen in years.

Very soon, we'll stop calling it "AI film" and just call it film.

[William Sousa @willw3b]

@Web3Kristel nice seats & capture! 🔥

[William Sousa @willw3b]

Higgsfield Soul Cinematic is underrated.

Barely anyone talks about it, but the images have a real cinematic mood.

The only BIG caveat:
You still can’t use an image ref and a prompt together, which makes precise direction harder than it should be.

I’ve always been a big fan of Wong Kar-Wai's films, so this is a small tribute while testing these models.

Image: Higgsfield Soul Cinematic
Video: Kling 3.0 4K

[William Sousa @willw3b]

AI compressed the difficulty of building faster than it compressed the difficulty of getting people to care.

Coding agents can solve a lot of build problems now.

But taste, timing, positioning, creativity, and human attention are still more art than science.

[Isha Parekh @iishaparekh]

marketing is 100x harder than programming

[William Sousa @willw3b]

if your brand context still lives in Google Docs, or you’re pasting it into Claude/ChatGPT every session, you’re behind.

the current standard is a shared repo with your context, agents, skills, workflows, outputs, and logs.

your brand should have its own operating system.

[William Sousa @willw3b]

@NoahKingJr the end

[William Sousa @willw3b]

New Higgsfield CLI now works on Cursor 🎉

Just tested out the new Happy Horse video model entirely within my Cursor session.

No browser needed, just typed the commands and saved on my local machine.

[Higgsfield AI 🧩 @higgsfield]

Meet Higgsfield CLI + Marketing Skills.

Instead of burning tokens on bloated schemas, or shipping broken creative at scale, the CLI keeps agent spend lean and Skills keep output high quality.

Pairs with Сodex, Claude Code, Openclaw etc.

npx skills add higgsfield-ai/skills

[William Sousa @willw3b]

@RoundtableSpace Outdated - kimi 2.6 should be there

[William Sousa @willw3b]

there’s a massive use case for Meta’s TRIBE v2 in growth and content marketing:

stress-testing creative before you spend budget distributing it.

[Alejo @ecommartinez]

🚨 ¡ESTO ES INCREÍBLE! 🚨

Meta acaba de lanzar TRIBE v2, una IA gratuita que predice exactamente dónde tu video se vuelve aburrido ¡ANTES de que lo publiques!

El 99% de los creadores no saben que esto existe. Así es como puedes aprovecharlo:

✅ El concepto: Un "espejo digital" que predice la respuesta del cerebro humano a tu contenido (video, audio o texto).
✅ La utilidad: Deja de adivinar. Detecta ganchos débiles, elimina los tiempos muertos y asegura un impacto máximo.
✅ El flujo de trabajo: Analiza la curva de respuesta, corta los segundos planos y empieza directamente con el resultado.

🔗 Prueba la demo aquí: https://t.co/dvLJkV3JLa