🤒 A common gripe in the medical field is that their early findings into potential cures for diseases are often mis-reported, giving those waiting for those life-saving ailments or medicinal breakthroughs unnecessary hope.

Is that about to happen again right here in this newsletter? We hope not but these latest claims are quite extraordinary. That’s because Google DeepMind CEO Demis Hassabis, one of leading figures in AI, is envisioning a future where Artificial Intelligence could cure all diseases within a decade.

But more on that later because, after yet another jam-packed week of AI news being vomited onto the masses, there’s a lot to cover in this edition…

🗞️ What we are covering today…

• Could AI cure all diseases within a decade?

• New startup Mechanize want to automate EVERYTHING

• OpenAI’s o3 model being deceptive about its capabilities

• Anthropic maps Claude’s morality

• UAE becomes the first country ever to use AI to write laws

• First AI makes Jesus a podcaster… now babies and dogs are too!

🔴 Quick Note: There’s a lot to cover this week, and for a better reading experience, we suggest opening this in your browser for the full experience! 

Head to the ‘READ IN BROWSER’ tab at the top of this email.

👁️ 👁️ What you might have missed

• Worried about AI taking your job and leaving you out of pocket? Well, a new startup Mechanize – led by Epoch AI co-founder Tamay Besiroglu – has announced its plans to replace human employees for AI bots, so this bastard is trying to make that a reality. Launched last week, Mechanize touted their frankly absurdly ambitious plans to “enable the full automation of the economy,” leading many on social media to get out their pitchforks. A traitor to humanity in the making?
  

• Open AI’s announcement of their o3 models last week was met with excitement, but users testing it out have met it with skepticism. That’s because Transluce AI tested a pre-released version of o3 and found it “frequently fabricated actions it never took”, before justifying these actions when confronted. These hallucinations are looking more like deception.

• Anthropic has released a groundbreaking study where they analyzed 300,000 conversations by Claude AI, mapping out its morality. The research saw Anthropic identify 3,307 ‘AI values’, helping align with the company’s ‘helpful, honest, and harmless’ constitution. It’s another stride to seeing AI fit in line with ethical guidelines, which is great. However, I’d hate to see someone map out half a million of my conversations and reveal that I’m an absolute idiot (something I already know.)
  

• Ever used AI to write you a cover letter or something you couldn’t be arsed to do? Yeah, me neither. But the United Arab Emirates have taken it a step further this week. Taking laziness to the next level, the UAE has become the first country in the world to use AI to write and update their laws. Government officials from the Gulf State claim this system will accelerate the legislative process by up to 70%, but I think we all know they really just couldn’t be bothered to do it themselves.
  

• Can you feel your heart pounding with excitement? Well, that’s because Lovable 2.0 has finally arrived! (Sorry for the shit pun). The Swedish software company has been edging us for quite some time now with the new update before finally releasing it on Thursday. Lovable 2.0 is now “smarter and more secure”, allowing you to build apps, tools and websites faster than ever.

🗣️ Other Titty Bits 

• Want to get yourself a meeting with investors? Just hack Calendly and schedule yourself in like this guy, simple

• California-based Play AI have released an AI voice changer, allowing you to use ANYONE else’s with little audio

• Out with Musk, In with Bezos: OpenAI now has Jeff Bezos on their side, after The Washington Post partnered with the AI company on search content. This messy break-up with Elon just got juicier

• Apple removes ‘available now’ banner from their Apple Intelligence page following a National Advertising Division (NAD) inquiry. Ouch

• Google Gemini has a whopping 350MILLION monthly users

• Donald Trump is set to sign an executive order that will integrate AI into K-12 schools

• Grok V 1.5 can now understand what is on the other side of your camera with its multimodal understanding rollout, Vision

• Perplexity comes for Siri and Alex with their new iOS Voice Assistant - Here’s how you can replace Apple Intelligence and Siri on your device along with a HUGE release week

• OpenAI’s brilliant 4o-image generation model is now available for developers globally to get their hands on. Expect to see racist uncles on Facebook Ghiblifying themselves real soon.

🦠 Could AI cure ALL diseases within a decade? Here's what DeepMind's CEO has to say

Ever wondered what the world would look like in 10 years time? With the rapid advancements in technology, medicine and science, it would be safe to assume anything is possible.

But for Google DeepMind CEO (Sir) Demis Hassabis, he believes we could be on the cusp of something that seems frankly unthinkable.

During his appearance on CBS’s 60 Minutes, Hassabis shared his bold vision that AI could effectively wipe out ALL diseases within a decade, which sounds more like science fiction than science fact.

Here’s what he had to say: “So on average, it takes 10 years and billions of dollars to design just one drug. We could maybe reduce that down from years to maybe months or maybe even weeks, which sounds incredible today, but that’s also what people used to think about protein structures.

“It would revolutionize human health and I think one day maybe we can cure all disease with the help of AI.

“I think that’s within reach. Maybe within the next decade or so. I don’t see why not.”

Demis isn’t some mug who is trying to drive stock prices through speculation and hyperbolic claims, which we see all too often with Elon. He was awarded the Nobel Prize in 2024 for his work on protein folding prediction models.

During that same interview, Hassabis didn’t stop there with making my little mind explode.

Hassabis claimed none of today’s systems feel self-aware or conscious at this current stage, but didn’t rule out it being theoretically possible in the next five years or so.

But what if these machines do become self-aware? Hassabis said we might not even recognise if they do.

So, to wrap up on this point, just try to stay alive for the next 10 years, alright?

📈 Trending tools, models and apps this week

📋 LLM Leaderboard

Gemini 2.5 Pro is still out in front with ChatGPT-o3 making up some serious ground since its release.

⚙️ Trending models

• Dia-1.6B: Turns any script into natural-sounding multi-voice dialogue with emotion in seconds.

• MAGI-1: Drop in a still image and get a smooth five-second video out.

• BitNet-b1.58-2B-4T: 2-billion-parameter LLM that runs on 1-bit weights, keeping costs (and temps) low.

• Flex 2 preview: 8B text-to-image with inpainting plus pose, depth and line controls baked in.

• GLM-4-32B-0414: 32B chat model built for heavyweight reasoning, coding and tool calls.

• HiDream-I1-Full: 17B image generator that nails photo-real, cartoon or concept art at blink-and-you-miss-it speed.

📲 Trending apps

• DeepSite: Describe a web idea, and it builds the whole thing for you, using DeepSeek.

• InstantCharacter: Upload a selfie and get yourself re-imagined in any art style.

• UNO FLUX: Mix text and reference pics to craft bespoke images.

• Kolors Virtual Try-On: See how clothes look on a photo before buying.

• EasyControl Ghibli: Give any scene the full Studio Ghibli glow-up with one prompt.

• FramePack Image-to-Video: Feed a still and get a five-second animated clip.

• Describe Anything: Point at part of a picture and it tells you what’s there.

• FLUX LoRa the Explorer: Load any LoRa, fire off fresh images and explore new looks.

• OmniControl Art: Restyle photos into Simpsons, Snoopy, Ghibli and more with a click.

• MotionShop2: Swap a video character with someone from your camera roll.

🙂‍↕️ Here’s everything we used this week 

We might have found an answer for “Fuck, I need to test this”, “Oh shit that looks interesting”, “hmmm I wonder what these models output and what is the difference”, “lemme save this for later”, “oh I should build this”… Does this sound familiar? 

Flowith 2.0 - dubbed an AI creation workspace, changes up the single chat interface and swaps it out for a flowchart. 

This allows you to compare different model outputs and continue to branch and change ideas all on a single canvas.

Say you want to plan for a new genius app idea that only you, on this planet of 8 billion people, have thought about.

You’re probably bouncing from AI to AI, agent to agent, messing around to see what works and then plugging outputs back in and it is all over the shop.

Flowith is different. It is all presented on one page in different branches, allowing you to keep track of your ideas.

Here’s a quick example of what it looks like on a page:

You can use different models, compare models, and utilise plugins for web search and video generation. Hopefully, you can see what I’m getting at here: Flowith gives you more creative control using AI from concept to final output.

They also have this thing called the “Knowledge Garden”.

One big thing at the moment with AI models or agents is the context that is given to them when you're trying to get to your end result. This is a store of any resources and data you collect from X, websites, and documentation, forming a personal database for use with Flowith.  

Here is a basic example of using the Big Machines Newsletter to create follow-up tweets for social media.

Yes, our newsletter is written by humans, but we will use AI in this instance to have a crack. 

I’ve asked Grok 3, GPT 4.1 and Claude 3.5 the same prompt:

“Please create a tweet based on the Big Machines latest newsletter I can use for Twitter/X, include key details on latest updates on LLMs and interesting stories as well as a hook to subscribe to the newsletter”.

One-shotted, however, you can see that it’s allowed me to explore each model side by side for the same prompt and also referenced the latest Big Machines Newsletter to create context for the output tweets.

Pretty simple and cool, and then I just progress forward with the best output!


🫵 Other cool products we found helpful this week (we hope you do too)

• Higgsfield AI is INSANEEEEEE. This new text-to-video model is absolutely nuts

• Apply for hundreds of LinkedIn jobs in seconds

• Change the sites’ ChatGPT sources for your research

• Pika Labs dropped a new iOS app that Pikaffect your videos

• Flora + GPT4o shows us more flowchart magic

🤝 In Partnership with Mira Network.

🤯 Did you know that AI models hallucinate, and around 30% of their outputs could be false?

This creates significant challenges for anyone relying on AI systems for critical tasks. Mira Network has developed a solution to this problem with its trustless verification network that fundamentally transforms AI reliability.

Their decentralized approach creates a new paradigm for AI verification, enabling users to confidently rely on AI outputs across multiple models (DeepSeek R1, GPT4o, and Llama 3. 70B Instruct as of now).

If you are a developer, Mira's Public Testnet is now live at console.mira.network 

The future of AI isn't just about powerful models—it's about verified ones.

Don’t trust; verify.

💸 Financials

• Ex-Meta engineer Boris Valkov has raised $14m for Lace AI, which seeks to be a… checks notes… “revenue-generating startup” - I am blown away that nobody thought of that earlier…
  

• Big Machines regular Mr Musk is looking to raise $20bn for xAI at a valuation that could make it the second largest capital raise of all time. Bloomberg are reporting an estimated valuation of $120bn.
  

• Supabase have nabbed a massive $200million Series D at a $2billion post-money valuation thanks to vibe coding. It comes just seven months after the open source database raised $80m led by Peak XV (a Sequoia spinoff) and David Sacks’ Craft Ventures.
  

• After mentioning this in our first Big Machines newsletter, it looks as if OpenAI would buy Google’s Chrome, according to a report. That browser is the keys to the AI agent kingdom and it looks as if this one is playing out. Our very own Grant hasn’t shut up about it for months - maybe he does know what he’s talking about?

• Manus (ANYBODY GOT AN INVITE CODE?!) has gone and raised $75m at a $500m valuation, a round led by Benchmark. Pretty big cheques continue to rain down on any and all products… maybe we should raise?

   

• Intel is reportedly planning to lay off over 21,000 employees – roughly 20% of its workforce – with an announcement expected soon. It’s barely a year on from when they announced 15,000 job cuts back in August 2024. Intel has been struggling for some time and it doesn’t look like it will improve any time soon, sadly.

  
  🤑 Other financial news 

• Datadog has acquired AI-powered observability startup Metaplane

• ListenLabs announces $27m fundraise, led by Sequoia

• Noxtua raises $92m for its sovereign AI, tuned for the German legal system

🕵️‍♂️ FREE ENTRY TO OUR INVITE-ONLY AI CHAT ON TELEGRAM…

If you share this newsletter with a friend and they actively sign up for the Big Machines newsletter, we will send you access to our invite-only Big Machines Telegram group, which is full of builders, investors, founders, and creators.

Access is now only granted to those who refer our newsletter to active subscribers, which means if you sign up on your work email, we will know you sneaky bastards.

This would kill our open rate, so please don't do that, we beg.

👋 Until next week

The great thing about the Big Machines newsletter is that it’s multi-layered, especially when it comes to memes.

If you recall back to our first newsletter, the one and only Jesus Christ tried his hand at podcasting (with the help of AI, of course), giving you a glimpse into what many label as their Lord and Saviour would look like if he appeared on The Joe Rogan Experience.

Then, last week, our second newsletter revealed how Google were literally trying to talk to dolphins, taking us a step closer to talking to our pets.

Now this week, it seems we’re getting a weird-as-fuck combination of the two (this time without Jesus). In a video posted on X, AI conjured up what a baby and dog would look like if they posted a podcast. See what I mean? MUL-TI-LAY-ERED.

Take a look for yourself:

Some other funny shit we found…

•  Gemini 2.5 Pro now has seven out of eight badges on Pokémon… this is the only benchmark we care about now.

• This guy is using ChatGPT to find underpriced items on Facebook Marketplace. A negotiation agent is definitely needed to put up with this cheap mfers over there. And NO we do not want to swap for a BMX.

• A tool that tracks how much energy YOU are directly using when you use AI… fml, JUST STOP OIL will be laying down in your front garden, stopping you from going to work if they find out.

• Signing off with the most terrifying message we have seen in quite some time… enjoy! The inevitable is coming. Be nice to the AI you use.

So, that’s another week done and dusted!

What do you think of Hassabis’ vision? Could all diseases be wiped out within the next decade? 

Let us know what you think and leave us some feedback on our newsletter to see what we’re doing right or wrong.

And if you’re feeling REALLY cool, share this newsletter with your pals – this newsletter’s success could be the difference between Sam having a hot meal or going hungry and you wouldn’t want to do that to him, would you?

Catch you next week and enjoy what’s left of your weekends.

Sam, Grant, Mike and The Big Machines team.