per my last email

per my last email

[zhod @zhodonx]

This is crazy.

Imagine being able to viberesearch a cure for HIV someday.

Anthropic & Citrini just used Claude to design proteins that could eventually help treat scorpion stings.

Here’s a recap:

Five people, three with zero science background, decided to see how far they could get using Claude to design a potential treatment for scorpion venom.

⟣ They let Claude coordinate hundreds of agents + specialist scientific tools.

⟣ the design campaign produced 211 proteins and used just over 10k H100 GPU hours.

⟣ the actual AI design process took about 64 hours

⟣ then scientists spent roughly two months synthesizing & testing everything in real labs

⟣ 17 of the designs successfully bound to their target toxins across 4 of the 5 toxins tested.

This is obviously nowhere near an actual antivenom, because there's no proof those proteins actually neutralize the venom.

But still.

I believe there’s insane potential in AI + science here, not just for industries or the wealthy, but for individuals.

Same way someone with no coding background can now vibecode a game, I can see AI doing something similar to scientific research.

Might sound like a stretch rn, but so did flying cars.

And this will only be possible if access don't eventually get hijacked by the govt.

We've entered the technological Singularity.

[Myndlab @Myndlab_ai]

We’ve rounded up the top 5 questions we received from visitors at AI Everything Abu Dhabi. 
Watch the video and COMMENT your questions below!

[Myndlab @Myndlab_ai]

A year ago you had the idea.

The booking app for your salon.
The Arabic reading app for your students.
A marketplace for your city.

This week, it could have its first user.

What's stopping you?

[Myndlab @Myndlab_ai]

We launched Myndlab at Ai Everything Abu Dhabi, and now we're bringing it to Dubai AI Festival!

Join us to discover how Myndlab is making AI-powered application development more accessible, secure and sovereign, empowering organisations to build locally and scale with confidence.

Come explore all you can build.

📅 26–27 October 2026  
📍 Dubai World Trade Center

AI App | Build With AI | Abu Dhabi Exhibitions | Tech Company | Build With Myndlab | Business Owners | Entrepreneurs | Build Your App

[Myndlab @Myndlab_ai]

Latest:

Anthropic just let outside safety testers work inside the company.

They even get to publish what they find, and Anthropic can’t veto it.

Not everyone agrees it’s truly independent.

But we believe it’s the right call.

Safety claims should be checked by outside teams who care most about the safety of the general public and how AI affects them.

This is a win for AI and sets an example to other frontier teams.

[zhod @zhodonx]

$137M has been spent on AI generated political Ads this year

In 2024 it was $50k for the whole cycle.

This year it’s over $137M.

⟣ 220+ campaign videos used AI fakery, & about 70% had no ai disclaimer.

⟣ one governor’s campaign faked the voices & faces of NYC’s mayor & NY’s governor.

⟣ 80%+ came from republican campaigns or aligned pacs.

Right now, it’s safe to say AI Deepfake Ads has taken over the US Midterms.

& at this rate the 2028 candidates won't even need to exist.

[zhod @zhodonx]

It looks like Europe is losing the AI race.

Mistral CEO Arthur Mensch says Europe leaving frontier AI to the US & China would be a fundamental mistake.

He wrote in his latest essay, "Build or surrender":

"for now, the choice is still ours."

Europe already pays over €260B a year to American cloud & software companies

And has slowly become "a continent that consumes what others build"

On a US lab cutting off its top models in June, he said nobody asked for Europe's opinion.

Mistral plans to grow its compute 6x next year

If you ask me, he's on a great track, but right now Europe's best model still ranks behind 7 Chinese open models.

[zhod @zhodonx]

I do think the biggest problem with fighting AI labs on training data is that most of it, according to them, is simply sitting on the open internet.

& so far, US judges have mostly called training fair use, and even Meta won after using pirated books.

So I'd say this fight gets won by labs, or simply forgotten at some point.

Or maybe it just gets bought out?

[zhod @zhodonx]

@ai_for_success Built this with Gemini argon last night.

Added a koi pond to the mix too

https://t.co/ysVWGz9IFK

[zhod @zhodonx]

Bill Gates: Rogue AI isn't the real danger rn.

"When you have bad people, no safeguards, and then the latest models…"

He expects "a lot of negative surprises" in cyber next year.

But I think we don't necessarily need the safeguards gone at this point.

A good example is the crew that ran 51 Claude Code agents in a campaign that stole 628k+ Ukrainian records.

& when Claude refused, they simply opened a fresh session calling it an "authorized pentest."

Chat, we are cooked.

[zhod @zhodonx]

Gemini 4 vs Opus 5.5.

I tried both models on the same prompt.

Each run took less than 17 minutes.

I’m leaning towards argon, but since both theme were so dark, I couldn’t really tell, but Gemini’s version looks well structured & has a cleaner feel.

Gonna run a few more prompts before calling it.

Which one do you prefer?

[Myndlab @Myndlab_ai]

Just in:

Epic Games, the company behind Fortnite, says it vibe coded its new launcher.

We’ve watched people use AI to build personal websites and tools for their businesses and teams.

As believers in building with AI, we see this as another step toward mainstream AI-assisted development.

Established software teams are putting these tools to work on products they already maintain.

It makes us curious about what gets rebuilt next.

Has your team shipped a product or internal tool using AI-assisted coding?

Tell us what it does.

[zhod @zhodonx]

I believe there’ll be an insane level of opportunity for creators in the coming months.

Especially around graduation airdrops on @vibevibefun.

Hear me out:

Builders can easily select creators for an allocation before launch. The airdrop depends on the project reaching graduation.

I can see a lot of them wanting creators who can help people understand what they’re building & bring the right audience to it.

Best part is you don’t need to buy in to be considered, although you still need to be selected.

As more projects launch, I expect creator allocations to become a bigger part of how builders grow their communities.

If you’re a creator, I’d recommend getting familiar with what’s going on here.

[zhod @zhodonx]

Make sure to give it a go:

https://t.co/vMYA87VVny

[zhod @zhodonx]

First look at Gemini 4 and o’boy it’s really good at 3D

Gemini really cooked.

Deepmind really went all in on making argon speak for itself even with all the fud & delays.

[zhod @zhodonx]

➥ 68 vs 67 on CWE-bench.
➥ 77.9 vs 74.2 on DeepSWE.
➥ 51.3 vs 42.5 on AutomationBench.

Gemini 4 Argon vs Opus 5.5, according to Google’s numbers.

To think this was the most ridiculed model on here for over a week after all those delays.

But what I’ve personally loved most about this isn’t just Argon’s release, but Google’s position at the frontier now.

I’ve always seen Google as a general purpose tech company trying to leave a mark on everything.

But with Argon here, I’m expecting a lot more from the Gemini 4 family going forward.

Still trailing Anthropic & OpenAI in some areas, but way more interesting than simply just reaching for their models on subagent tasks.

There’s a good future for deepmind here.

[zhod @zhodonx]

Another Gemini cook.

Showed Argon the DeepMind pagoda demo & asked it to build a better interactive version I could walk around in.

I never suggested features. I just kept asking it to recommend improvements over a few iterations.

It added the controls, koi pond & interactions, and ringing the bell on its own.

About 4 hours, my entire week's usage on Devin.

What do you guys think?

[zhod @zhodonx]

Yea, this is impressive @bot

I had this conversation last week Monday & we somehow could not find a satisfactory resolution for it, which made me skip it.

Fast forward to today, it just pulled it up.

PS:
I did not schedule the task or anything.

[zhod @zhodonx]

Jacob Coxon: “I am a believer in the future utopia.”

AI does the work, and the economy runs on automated production, then everyone gets a share of the money.

“there will be no jobs.”

That’s the future he says these companies are building toward.

Jon Stewart on the AI utopia:

“They’ll build systems to create food, then give us money, and then we’ll go to that place and give it money… It’s basically just a circle jerk.”

And coxon also says their incentives are concentrating power in fewer hands.

What drives them day to day is getting there first.

So basically, the wealth, status and importance that comes with it.

Jon Stewart then asks whether the utopian promise is a cover for naked ambition.

Jacob says both are mixed together:

“the naked ambition is kind of the important bit.”

I personally think this is BS btw.

A handful of people get to reshape the economy, decide what your future looks like, and hand you money to live in it.

[zhod @zhodonx]

I hate to bring the open-weight vs closed lab convo back on the TL.

But what exactly happens when every region starts wanting their own AI?

I personally got struck with this after the Mistral Large 4 release.

I’ve seen a lot of backlash over its performance, but I don’t think the team is trying to smoke the frontier at anything rn, but instead, pushing pretty hard for European AI sovereignty.

And Europe isn’t alone on this.

South Korea, Japan, the UAE, UK and a bunch of other countries are already putting serious resources into their own sovereign AI projects.

So naturally, I start to wonder how much more regional this could get.

Because obviously, not every country can afford to train a frontier model from scratch.

Most probably can’t.

But I can imagine more governments wanting AI they can run themselves, adapt to their own laws, and use without depending entirely on US or Chinese labs.

And this is where open weights make a shit ton of sense.

Countries can build around them and have much more control over deployment.

Meanwhile, closed labs would probably have to offer more local infra and flexibility if they want to keep those governments as customers.

And if Europe keeps putting serious compute, money and research behind its own models, who’s to say what the gap looks like a couple generations from now?

Especially when some governments might happily accept a slightly weaker model just to have more control over it.

I reckon this whole sovereign AI push could make more sense in the future

Thoughts?