Applied AI Engineer • DevRel 🥑 DM for Work / Paid Collabs ☕ I build Custom Voice AI & Agents for businesses Shipping AI Products in Public @debugginglife25
vishalsinghpanwar02@gmail.com
Joined April 2023
- Tweets2.8K
- Following720
- Followers14.9K
- Likes49K
Pinned Tweet
WTF 🤯
Imagine catching hidden security bugs in apps like WhatsApp or banking apps before hackers ever find them.
Now imagine doing the same for hospital machines, cars, and even power grids.
Google just took a HUGE step toward making this possible. They used AI to mathematically verify real encryption software in just 4 weeks 🔥
Lemme explain why this is BIG, in simple words 👇
Think about how much we trust software every day.
We send private messages, transfer money, store personal information, and just expect everything to be secure.
But how do we actually know?
Normally, developers test their code, look for bugs, and try different ways to break it.
The problem is that even if thousands of tests pass, there could still be some hidden mistake nobody thought to check.
There's another way to check software called formal verification.
Basically, instead of just testing whether something works, you use math to PROVE that certain things can't go wrong.
Sounds great, right?
The problem is that doing this for complicated real-world software is insanely difficult. It can take experts years.
And this is where DeepMind's research gets interesting.
They picked Signal, a private messaging app similar to WhatsApp, where encryption helps keep conversations private.
In fact, versions of the Signal protocol also help protect messages in WhatsApp and Google Messages.
Now, researchers have spent decades studying how to make encryption secure.
But proving that encryption is secure on paper is one thing. Proving that the actual code follows those same rules is much harder.
Because even if the design is perfect, one tiny coding mistake could still create a security hole.
So, DeepMind built something called AlphaProof Code to tackle that gap.
First, human researchers worked with AI to define exactly what security protections they wanted to prove.
Then teams of AI agents powered by Gemini worked together to figure out the mathematical proofs.
And here's the crazy part.
The agents generated around 6.5 MILLION lines of proof code 😭
But they didn't just trust whatever the AI came up with.
Every proof had to pass through Lean, a separate mathematical checker that verifies whether the reasoning is actually valid.
So, the AI couldn't just confidently say "yep, this is secure" and move on lol.
And they went all the way down to machine code, the actual instructions a processor executes.
So, what did they actually prove?
Imagine someone manages to steal your current encryption keys.
That shouldn't automatically let them go back and read all your older private messages.
And if someone temporarily compromises your device, the encryption system has ways to make future conversations secure again once the attacker loses access and fresh keys are exchanged.
It also needs to make sure messages really come from the person they're supposed to come from.
DeepMind mathematically verified important protections like these in Signal's compiled encryption code, under specific assumptions.
And one experiment really caught my attention.
They took 2 real security bugs that had already been discovered and fixed in Signal, then deliberately put them back into the code.
In both cases, their system mathematically proved that the buggy versions broke the expected security guarantees.
That's pretty wild.
Instead of just looking for suspicious code, they could actually show when important security protections were broken.
Now, this doesn't mean Signal is suddenly impossible to hack. They verified specific parts of its one-to-one encryption, not every possible security issue in the entire app.
But think about what happens if we can eventually use this approach for other software.
Banks could verify important protections in the systems handling our money.
Hospitals could check whether medical devices follow critical safety rules.
Car manufacturers could verify software controlling braking and other safety features.
Even power grids and aircraft could get stronger guarantees about the code keeping critical systems running.
These are all places where one tiny software mistake could have serious consequences.
And right now, getting this level of verification is so complicated and expensive that most software never goes through it.
If AI can make it much faster and cheaper, we could start checking systems that would've been too difficult or costly to verify before.
Imagine catching certain serious mistakes before software ever reaches users, instead of discovering them after something goes wrong.
And the timing of this is what makes it even more interesting to me.
AI is already getting ridiculously good at writing code.
We're moving toward a world where AI agents can build entire apps and complex systems much faster than before.
But who's going to check all that code?
Because writing software faster doesn't automatically mean we're building safer software.
And that's why I find this research so exciting.
We've spent years trying to make AI better at writing code. Now we're seeing it help mathematically verify that important parts of that code actually work the way they're supposed to.
Of course, DeepMind built on years of earlier verification research, and there's still plenty of work needed before this becomes practical for every kind of software.
But doing this on real, widely used cryptographic code in just a month is still pretty insane.
Imagine a future where the apps and systems we depend on don't just come with thousands of passing tests, but actual mathematical proofs for their most important protections.
We might be heading toward a world where AI doesn't just help us build software faster.
It helps us prove we can trust what we've built too.
And honestly, that could end up being even more important.
What a time to be alive 🔥
Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified – by any method, with or without AI – and at this depth.
More than a research breakthrough, it’s a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats.
Formal verification offers a higher standard of security; however, it’s extremely complex and time consuming, which is why in many cases it’s considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month’s time.
Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices. This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work.
Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work.
On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary.
I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger, Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership.
deepmind.google/blog/formal-…
dumb me realizing i can literally take a screenshot of any tweet directly from the copy link option
1
58
So now I can build AI agents that make quick decisions without waiting for an LLM to generate a whole response.
Can check up to 1,000 things in one call, pick the right tool or model, and get confidence scores. No output tokens needed.
Basically, fewer AI calls, faster agents, lower costs.
Nice work, Google team 🔥
grok bot 1.0 just dropped 🔥
you no longer have to tell your AI what to do every single time.
bots can now:
• join team chats & calls
• remember context and decisions
• spot unfinished work and follow up
• handle research, docs, slides & workflows
love this. we're officially entering the superintelligence era 🥳
WTF 🤯
We might be closer to understanding how to treat some of the world's toughest diseases than most people realize.
Google just unveiled AlphaGenome 2, an AI that can predict how changing a SINGLE letter in your DNA affects different cells in your body.
And in one major test, it identified nearly TWICE as many important genetic variants as its predecessor.
This could be a huge step toward understanding what actually drives diseases and finding better treatments.
Lemme explain why this is BIG, in simple words 👇
First, here's something pretty wild.
Your brain cells, heart cells, immune cells, and almost every other cell in your body contain basically the SAME DNA.
Yet they all do completely different things.
How??
Think of DNA as a giant instruction manual.
Every cell gets the same manual, but each one reads different chapters.
Your heart cells follow instructions that help your heart beat.
Your immune cells follow instructions that help fight infections.
And your brain cells use instructions that help them send signals and process information.
Now, DNA doesn't just contain genes.
It also contains regulatory regions that work kinda like switches, controlling which genes turn on, when they turn on, and how active they become.
Sometimes, changing just ONE letter in your DNA can mess with one of those switches.
Here's where things get interesting.
The exact same mutation might cause problems in one type of cell while doing almost nothing in another.
It could even behave differently in the SAME cell depending on whether that cell is resting or fighting an infection.
Same DNA. Same mutation. Different effects.
And this is one of the biggest challenges in understanding genetic diseases.
Scientists have gotten really good at finding DNA changes associated with diseases.
But knowing a mutation is linked to a disease doesn't automatically tell you WHY.
Which gene is it affecting?
Which cells are involved?
Is it making that gene more active or less active?
And how does that contribute to the disease?
Answering these questions can take years of complicated experiments.
This is where AlphaGenome 2 comes in.
Google's original AlphaGenome could analyze around ONE MILLION DNA letters and predict different kinds of molecular activity.
Pretty impressive already.
But much of its training data came from whole tissues and lab-grown cell lines.
And here's the problem.
A tissue contains lots of different types of cells.
Take your gut, for example.
Some cells absorb nutrients.
Others protect the gut lining.
And some are immune cells fighting infections.
When scientists measure an entire tissue, they're often looking at an average of what all those cells are doing.
But what if a mutation only affects ONE specific kind of immune cell?
That effect could get buried in the average.
It's like trying to figure out which student is struggling when all you can see is the class average.
So Google redesigned the model.
Instead of looking mainly at broad tissues, AlphaGenome 2 can predict DNA activity in specific cell types and even particular cell states.
Researchers can ask:
What happens if this DNA letter changes in a brain cell?
What about an immune cell?
What if that immune cell has just been activated by a vaccine?
And the model can predict how the same genetic change might behave differently in each situation.
To make this possible, Google used data from more than 107 MILLION individual cell profiles across 68 tissues, alongside other genomic data.
That's an insane amount of biological information 😭
But how much better is it actually?
Google tested AlphaGenome 2 on a dataset covering 28 types of human immune cells.
At a strict 90% precision threshold, the previous model, even after being adapted to single-cell data, identified 16% of the tested likely causal genetic variants.
AlphaGenome 2 identified 30%.
And when predicting whether variants increase or decrease gene activity, its coverage at 90% accuracy jumped from 23% to 43%.
Nearly TWICE as many important effects identified at those high-confidence thresholds.
But the example that really caught my attention involved inflammatory bowel disease.
It's a condition where parts of your digestive system become chronically inflamed.
Researchers had previously discovered that a particular DNA variant can increase the activity of a gene called ETS2, which plays an important role in inflammation.
Here's the crazy part.
The mutation wasn't even inside that gene.
It was in a regulatory region hundreds of thousands of DNA letters away.
And it mainly affected macrophages, a type of immune cell found in the gut.
So Google tested whether AlphaGenome 2 could reproduce this already-known mechanism.
And the model singled out the exact variant.
It predicted that the mutation would increase ETS2 activity specifically in the relevant gut macrophages, while leaving other cell types and genes largely unaffected.
That's actually wild.
One tiny DNA change.
A gene located far away.
A complicated inflammatory disease.
And the AI connected the variant to the right gene AND the relevant cells.
Now, this wasn't a brand-new disease discovery. Scientists had already established this mechanism experimentally.
But the fact that AlphaGenome 2 could reproduce it computationally is what makes this exciting.
Imagine using this to investigate genetic changes scientists DON'T understand yet.
Instead of testing countless possibilities, researchers could narrow down which variants, genes, and cells deserve attention first.
And there's another upgrade that's pretty interesting.
AlphaGenome 2 can adapt to new cell states without retraining the entire model.
Researchers can give it gene-activity measurements from just a few dozen cells, and it can use that information to make predictions for those cells.
Google demonstrated this using data from a COVID-19 vaccination study.
The model distinguished resting immune T cells from activated ones.
And it predicted that a genetic variant associated with severe COVID-19 had a stronger effect on gene regulation when those cells were activated.
Think about that.
A mutation might have a relatively small effect when an immune cell is resting, but a much stronger one when that same cell becomes active.
So studying the wrong cell state could mean missing something important.
And this is where things get REALLY interesting.
Researchers also tested the model on congenital hyperinsulinism, a rare condition where the body releases too much insulin.
Certain DNA changes can cause a gene called HK1 to become abnormally active in insulin-producing pancreatic cells.
AlphaGenome 2 predicted these effects specifically in pancreatic beta cells rather than broadly across unrelated cell types.
Again, it reproduced an already-known disease mechanism.
But imagine applying this to rare genetic disorders whose underlying mechanisms are still poorly understood.
Or Alzheimer's disease.
Scientists have identified genetic variants associated with increased risk, but figuring out exactly which brain cells those variants affect and which genes they regulate is much harder.
A model like this could help narrow down the most promising explanations.
The same applies to autoimmune diseases, where the immune system mistakenly attacks the body's own tissues.
Understanding how specific genetic changes affect particular immune cells could reveal biological mechanisms that researchers might otherwise overlook.
And then there's drug discovery.
This might be one of the biggest long-term implications.
Finding a gene linked to a disease doesn't automatically tell scientists how to treat it.
They need to understand what that gene is doing, which cells are involved, and whether changing its activity might actually help.
If AI can help answer those questions more accurately, researchers could potentially identify promising drug targets that might otherwise be overlooked.
But finding a drug target isn't the same as finding a cure.
Scientists still need to develop medicines, test their safety, and establish whether they actually work.
Understanding what's going wrong is just an important first step.
Google has also generated predictions for nearly 7 MILLION genetic variants across hundreds of human cell and tissue conditions.
Basically, researchers now have an enormous map of predicted genetic effects they can explore.
And AlphaGenome 2 can even estimate certain missing molecular measurements using biological information already available.
This could help scientists decide which expensive laboratory experiments are actually worth doing next.
The model is also available through an API for non-commercial academic research, allowing researchers to explore and test its capabilities.
Now obviously, this doesn't mean we're suddenly going to cure cancer, Alzheimer's, or every genetic disease tomorrow lol.
These are still AI predictions, not proof.
The model can miss important genetic effects, and scientists still need real experiments to confirm what it predicts.
It's not approved for medical diagnosis either.
But think about the bigger picture.
For decades, we've been getting better and better at reading DNA.
We've sequenced genomes, discovered millions of genetic differences, and found countless links between those differences and diseases.
But there's always been this huge gap between knowing that a mutation is associated with a disease and understanding what it's actually DOING inside the body.
That's the gap AlphaGenome 2 is helping close.
We're moving from:
"This part of your DNA is linked to this disease."
Toward:
"This specific DNA change might affect THIS gene, in THESE cells, in THIS particular way."
And that's a completely different level of understanding.
Imagine a future where researchers can investigate the predicted effects of a mysterious mutation across hundreds of cell types.
They could narrow down which cells are most affected, which genes might be involved, and which biological mechanisms deserve further investigation.
Instead of spending years figuring out where to look, they could potentially get to the most promising experiments much faster.
That's what excites me about this.
Not that AI has suddenly found cures.
But that it's becoming a powerful tool for understanding the biology behind diseases.
Think about how much biological data we're collecting right now.
Researchers around the world are mapping the human body cell by cell.
We're learning what different cells do, how they change, and how they behave during disease.
And now AI is getting better at connecting all that information to the genetic changes that influence those cells.
We've gone from discovering the structure of DNA, to sequencing the human genome, to identifying mutations associated with diseases.
And now we're building AI models that can predict how those mutations affect specific cells.
Not perfectly.
Not for every disease.
And definitely not without more experiments.
But the progress is real.
We've spent decades learning how to READ the code of life.
Now we're getting better at predicting how different cells actually USE that code.
And if we keep improving our understanding of what goes wrong inside our cells, we might get a whole lot better at figuring out how to fix it.
What a time to be alive 🔥
AlphaGenome 2 is helping uncover the cellular drivers of disease, giving scientists a powerful tool to predict how DNA variants impact specific cell types.
Every cell in our body shares the same DNA blueprint, but different cell types (like heart muscle cells vs. immune cells) read that blueprint differently. This new @GoogleDeepMind model predicts how these specialized cells interpret DNA, turning genetic code into biological insights.
Building on AlphaGenome 1 this makes it significantly easier for researchers to link genetic findings to the root causes of disease in the specific cells where they actually unfold.
Gemini Carbon is already out.
WTF is this now....some other release?
Hydrogen? Oxygen?
@TokenGremlin 👀
Why will AI agents need a completely new kind of web search, instead of just using Google 🤔
Parag Agrawal had a really interesting explanation for this.
We've spent decades building search around how humans use the internet, but agents behave so differently that the whole infra might need to change.
Think about how we use Google. We type a few random words, get some links, click around, read a bit, go back, search again. Half the time we don't even know exactly what we're looking for until we start browsing.
That's what search engines have spent decades optimizing for.
But agents don't work like that.
An agent can send an entire task's worth of context in a single query. And it doesn't need 10 blue links to click through. It needs useful information it can put into its context window, reason over, and figure out what to do next.
So, both the input and output of search change.
Then there's the scale problem.
If agents end up using the web 1000x more than humans, are we really going to build 1000x the search infrastructure Google has today?
Probably not. We'd need to rethink how search itself is built.
And the timing is interesting too. GPUs are getting more capable, and we're getting better at distilling intelligence into smaller models. So, building efficient search and ranking systems in entirely new ways is becoming possible.
Then there's latency, which I hadn't thought about either.
For us, a Google search taking 500ms feels fine. But for a voice agent, that same delay can feel awkward. You'd want something closer to 50–200ms.
Meanwhile, an agent researching something overnight could spend an hour searching. Nobody's waiting for it.
So even what counts as "fast" depends on what the agent is doing.
And the economics change too. Human search makes enough money per query to support its infrastructure. Agent searches could happen far more frequently while bringing in much less revenue per query. The same compute costs might not make sense anymore.
That's what made the whole thing click for me.
It's not just that agents will search more. They ask differently, consume information differently, and need different tradeoffs between speed, cost, and compute.
We've spent decades building search for humans clicking through websites. Now agents might become some of the biggest users of the web, and they don't even interact with it the same way we do lol
Makes me wonder how many other things we've built around human behavior that we'll need to rethink once agents become the users.
CRAZY TIMES AHEAD 🔥
i so wish I had the Gemini Ultra plan rn😭
feeling major FOMO about building with Carbon and seeing for myself if it actually delivers 👀🔥
Even if Gemini Carbon drops today, what's even the point? Fable 5.2 has already mogged it.
I genuinely don't get why Google keeps doing this to itself 😪
FUCK another AI company got hacked 🤯
and apparently hackers are now stealing API keys, selling AI access, and leaving someone else with the compute bill.
and TIL this is actually a thing called LLMJacking 💀
hackers got in through a cloud provider and managed to access customer credentials, billing info, datasets, and even incomplete model weights.
kinda same as crypto theft.
except here, instead of draining your wallet, they steal your AI access and make money off it.
Fireworks' CEO thinks the attackers used AI to speed up the hack.
if this is what hackers are already doing with AI, imagine what's coming next 😭
lots for us to learn and improve in cybersecurity imo.
Fuck, not GPT-6.2 again 😭 but this is actually important.
OpenAI just built something called Control Room. Basically a kill switch for AI research.
Lemme quickly explain why this matters 👇
OpenAI runs tons of experiments where AI models are trained, tested, and given tasks to do on their own.
Now imagine one of those AI agents starts doing something dangerous. Who stops it? And how fast?
That's what they built this for.
Every research job now gets an ID. Whenever it asks a model to generate something, the system checks if that job is still allowed to run.
If a separate AI monitoring it spots something serious and a shutdown is triggered, Control Room blocks further model output within 35 seconds.
And even if Control Room itself crashes or someone tries to disable it, the system is designed to block requests rather than let things keep running unchecked.
It's already being used across OpenAI's research infrastructure.
And this matters because we're giving AI agents more freedom to use tools, make decisions, and work for hours without someone watching every move.
Spotting dangerous behavior is one thing. Actually being able to stop it is another.
We're all focused on how much smarter AI is getting, but making sure we can still shut it down when something goes wrong is just as important today 🔥
We built Control Room, a central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.
It connects every research inference workload, including agentic sampling for model training and evaluation, data processing, and more. This works together with our misalignment monitoring systems, and halts model output within 35 seconds of a decision to intervene to stop or pause a job.
Control Room is just one part of our holistic approach for safe frontier model training and deployment.
openai.com/index/control-roo…
WTF… Claude just did some insane stuff across engineering, chemistry, math and physics 🤯
lemme quickly explain what they actually did 👇:
• Hardware: Gave Claude a simple prompt and it designed 3 working electronic devices from scratch. Circuit boards, components, software, 3D-printed cases, everything. Engineers built them and all 3 worked without circuit fixes. Basically turned an idea into real hardware.
• Chemistry: Showed it pictures of 4 complex molecules and asked how to make them. No internet, no research papers. Claude figured out full chemical synthesis plans, checked its own work and flagged risky steps. Stanford, Yale and Caltech professors rated the plans around grad-student level. Not lab-tested yet, but imagine how useful this could be for research.
• Math: Took on 203 unsolved optimization problems and beat the best-known solutions in 72 cases. Even proved optimality in 45. One involved scheduling 42 trains, where it found a way to cut peak power demand by 2.6%. Think logistics, transport, energy grids, all that.
• Physics: Worked on 2 open problems involving black holes and quantum gravity. One was a calculation left unfinished since 2003. Claude found an answer and also figured out an exact mathematical limit researchers had only been able to estimate for years. Both are now in preprints.
and the crazy part is these are completely different fields.
we're going from AI answering questions to AI actually designing things, helping plan experiments and tackling research problems people have been stuck on for years.
some results still need real-world testing and independent verification, but still man…
what a time to be alive 🔥
WTF 😭
first time in AI history we're seeing an older model STILL beating the new one on a benchmark LMAOO 💀
anthropic really put that in their own chart 😂
gotta respect the honesty tho ❤️
WTF Meta is coming for the AI model market 💀
they just launched Muse Ember, matching Claude Haiku 5.5 in intelligence at 1/3 the cost per task ($0.14 vs $0.41)
for developers building AI agents that make thousands of tool calls a day, this could cut costs massively.
huge win for startups building AI products on tight budgets.
Releasing Muse Ember today -- our smallest model that delivers higher intelligence per dollar than others in its class. Built for token-heavy work that agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing. Available today on the Meta Model API and in Muse Code.
Jensen Huang fucked up and Google is fully using this once in a lifetime opportunity lol 💀
Google just announced TWO new Nvidia-rival AI chips, with 30% cheaper TPUs for comparable workloads. Timing literally couldn't get better 😭
Bro this could reshuffle HUNDREDS OF BILLIONS across AI, and I don't think people realize how big the domino effect could be.
Lemme explain 🔥
$NVDA lost around $335B in market cap at one point today amid Jensen's Chinese spy-linked controversy. No criminal charges yet but if this becomes a leadership crisis, things could get ugly.
Meanwhile, here's what Google just unveiled:
• TPU 8tx → Secure AI for banks, governments and enterprises. Let's them run and train AI agents on sensitive data without exposing it, with up to 2x lower latency than TPU 8t.
• TPU 8i Max → Double the compute and memory bandwidth of TPU 8i. Built for AI agents handling entire codebases, hours of video and million-token contexts.
And Google already has nearly 100 TPU customers.
Apple and Anthropic are onboard, Oracle is bringing TPUs to its cloud, Meta is working on TPU compatibility, and OpenAI has reportedly explored Google's chips.
These companies spend BILLIONS on AI compute. They don't even need to abandon Nvidia. Just shifting a fraction of future orders could change where all that money flows.
Now look who benefits 💸
• $GOOGL → More TPU sales, cloud revenue, cheaper Gemini compute
• $AVGO → More custom chip orders
• $TSM → More chip manufacturing demand
• $AMD → Its EPYC CPUs already power Google's new 8tx systems
• AI giants → More choices, lower potential costs, stronger bargaining power
• AI startups → Cheaper APIs, better margins, more AI agents
• Banks & governments → Secure AI on sensitive data
• $NVDA → Potential pricing pressure and customers diversifying
The funniest part?
Google's been building TPUs for over a DECADE. They were already coming for Nvidia's AI billions.
And now Jensen's mess gave them the perfect opportunity to make their move lol 😅
Today, we're expanding our eighth-generation TPU portfolio with two new systems built specifically for the agentic era — bringing our total to four distinct eighth-generation TPU systems launched in a single year.
TPU 8i Max was built for massive context windows and the highest compute density per rack, and TPU 8tx is optimized for secure and confidential inference. While flexible across workloads, the four systems in our 8th gen TPU portfolio give developers and organizations a purpose-built accelerator for every stage of the agentic lifecycle.
@AminVahdat, SVP and Chief Technologist, AI and Infrastructure, explains why we're expanding our TPU portfolio and what everyone needs to know about TPU 8tx and 8i Max ⬇️
Apple is finally turning Siri into a proper AI agent 🥳
Siri can already understand your screen, emails, messages & personal context.
But Apple is now taking things further 🔥
Soon, Siri could:
• Handle complex, multi-step tasks
• Book restaurants, place orders & track deliveries
• Get things done across multiple apps with one request
• Work more deeply with your calendar, files, photos & messages
• Turn your iPhone, Mac & apps into one connected AI experience
basically, Apple's answer to Grok Bot, Hermes, OpenAI Dots and Claude, but baked into the entire Apple ecosystem.
sure, Apple is late to the party. but with Apple, you never know.
they were late to foldables too. just look at the iPhone Duo lol
Holy Moly 🤯
DeepSeek Harness launched on August 13th. By November 4th, it had already reached 1.54 million daily active users, consuming 92 trillion tokens a day.
DSH will win 2027.
Open AI will win 2027.
