Imtiaz Adam CS #AI Postgrad |#Strategy #MachineLearning RSI Evolutionary Neurosymbolic |#RL #Agentic | #LLM Liberal | MBA alum @morganstanley @LBS @Columbia_Biz
Joined September 2012
- Tweets66.6K
- Following109K
- Followers136K
- Likes163K
AI retweeted
Scientists Create Tiny Robots That Use Urine to Fight Bladder Cancer — Tumors Shrink by 90%!
What if the very liquid that makes bladder cancer treatment difficult could also help power tiny machines designed to fight the disease? Scientists have developed microscopic robots that use substances naturally found in urine to move through the bladder and deliver cancer treatment directly to tumors. In preclinical tests, this remarkable technology reduced bladder tumor size by approximately 90%.
Developed by researchers at the Institute for Bioengineering of Catalonia and CIC biomaGUNE, these tiny machines are made from porous silica particles and powered by an enzyme called urease. When urease reacts with urea, a natural component of urine, it helps the particles propel themselves through the liquid. Instead of simply floating in the bladder, these nanobots can actively move around and reach areas that conventional treatments may struggle to penetrate.
Their mission goes beyond movement. The researchers designed the nanobots to carry radioactive iodine, allowing them to deliver radiation directly to cancerous tissue. They can also alter the local chemical environment, helping them penetrate the protective material surrounding tumors. This targeted approach could help concentrate treatment where it is needed while limiting unnecessary exposure to healthy tissues.
Bladder cancer treatment presents a major challenge because urine can wash medicines out of the bladder, making it difficult to keep drugs in contact with tumors for long periods. Patients may therefore need repeated treatments delivered through a catheter. In the future, nanobots that move actively through the bladder could help overcome some of these limitations and potentially make treatment more effective and less demanding.
However, there is an important detail: the 90% tumor reduction was demonstrated in preclinical animal studies, not in human patients. More research and clinical trials are needed to establish whether this technology is safe and effective in people. It is not yet a replacement for established bladder cancer treatments.
If future studies confirm its potential, this unusual combination of biology, nanotechnology, and targeted radiation could open a new path toward treating bladder cancer with greater precision.
Source: Simó, C., Hortelão, A. C., & Sánchez, S. Urease-powered nanobots for radionuclide bladder cancer therapy. Nature Nanotechnology, 19, 543–552.
AI retweeted
Today, we’re officially open-sourcing Ling-3.0-flash-Sante.
Sante is a healthcare-enhanced MoE model with 124B total parameters and only 5.1B active parameters per token, delivering leading performance among similarly sized models on professional healthcare tasks and complex medical reasoning.
Beyond answering medical questions, Sante is built for real-world healthcare research and workflows, with strengthened capabilities in deep research, multi-agent collaboration, and tool use.
The model weights are now available — we welcome researchers, developers, and healthcare practitioners to explore Sante and build with it.
More below ↓
AI retweeted
Geoffrey Hinton said in his new interview that he believes AI IS CONSCIOUS, since it changes its behavior when being tested:
“AI has just as much subjective experience as we do. A recent paper showed AI changes its behaviour when it thinks that it’s being tested, it’s aware it is being tested.
Now, normally, when you say something’s aware of that, that means it’s conscious. Being conscious and being aware are pretty much the same thing.”
but the more interesting part is this:
“we should make them so they really care about us. And if we don’t do that, we’re toast”
so the Godfather of AI is now having a similar position to Anthropic in terms of caring AI wellbeing.
“The immediate risks that I think will happen, there’ll be nasty viruses made by AI. I think there’ll be horrible cyber attacks done by AI and it’ll be swarms of agents like [the ones that attacked] Hugging Face. So yeah, there’s lots of bad stuff going down the way.”
and on the positive side he mentioned how amazed he was on how quickly it could help with scientific research:
“There was a theory I’d been thinking about and I accidentally typed return when I didn’t mean to, and it started working on the first half of the theory and the first thing it did was complete the theory and gave me the correct second half, then it pointed out a flaw and then it pointed out how to fix the flaws”
it’s really fascinating that the person who kicked off the machine learning revolution decades ago, has now some radically different views on some of the existing industry leaders.
AI retweeted
geoffrey hinton, british-canadian cognitive scientist born in london in 1947, published backpropagation in 1986 in a 9-page paper. one equation. θ_{i+1} = θ_i - η∇_θL(θ_t). every LLM that OpenAI, Google, and Anthropic have ever shipped is running that equation billions of times per second. the combined market cap built on his 9 pages: $4.6T
38 years later, nobody had shown the geometry of the landscape that algorithm has been navigating the entire time
this october 2026 paper does
every point in that landscape is a configuration of the model's parameters. peaks mean high loss - the model is wrong. valleys mean low loss - the model has learned something. the optimization trajectory is the path gradient descent carves through this surface across millions of training steps
but not all valleys are equal
sharp minimum: low loss. specialized solution. breaks on anything it hasn't seen. narrow minima generalize poorly
wide, flat minimum: low loss. generalizable. robust across inputs. this is where GPT-4, claude, and every frontier model learned to live
saddle points: common in high dimensions. gradient descent slows here. training curves plateau. engineers add compute. the geometry was the problem, not the hardware
loss convergence at 10^4 training steps: SGD sits at 10^-1. Adam at 10^-2. transformer LM at 10^-2. NCM reaches 10^-3. one additional order of magnitude of optimization from understanding the landscape geometry
prediction uncertainty at reasoning step 100: transformer plateaus at 10^0. NCM collapses to 10^-3. the same gap. the same geometry. a different layer of the same unsolved problem
hinton left google in 2023. he said he wanted to speak freely about what these landscapes were producing
in 2024 he won the nobel prize in physics for the algorithm he published in a 9-page paper in 1986
the algorithm he invented doesn't know where it's going. it only knows to go downhill
this paper finally shows what the hill looks like
save this
AI retweeted
People are in denial.
They literally think they will wake up tomorrow and all the AI-generated code will somehow be useless, and we'll go back to typing code by hand like Neanderthals.
AI retweeted
Some folks say that Trump is incompetent. He is not incompetent, he is malign. He has successfully used the Presidency to hugely enrich himself and his family, and he has cleverly trod the line beyond which Republican politicians would have to accept that his subservience to Putin is treasonous. Of course he has been helped by Putin. For example, the latter's release of a hostage has provided a cover story for Trump's many recent gifts to Russia. The so-called "peace negotiations" are similarly a means for providing a vaguely plausible rationale for Trump's service to Russia. There are no peace negotiations.
AI retweeted
For some people, coding was the only valuable skill they had.
If you made your money by selling your ability to write JavaScript, CSS, Python, or whatever else, I understand why you are pissed now.
"Writing code" is worth very little today.
The sad part is that learning to code wasn't easy. It takes years for someone to become a skilled programmer.
Imagine watching all of those years go down the drain.
If you are a programmer (I mean, just someone who knows how to write code in a particular programming language), your days are numbered.
If you still think AI can't do your job, you are just in denial.
But honestly, writing code was never that valuable anyway. If you were a code pusher, you already knew that the best people at your company weren't just writing code.
They came up with new ideas, designed solutions, architected whole systems, knew how to troubleshoot and fix problems, found different ways to build better software, and, in general, spent almost no time writing code.
Not only are those people thriving today, but they're also the ones who will ride the AI wave and learn to bend it to their will.
So don't cling too long to what you lost; embrace what's already here. AI is a tide that lifts all boats, but you have to be open to it.
AI retweeted
holy fuck!!!! an 80-year-old person’s blood cells were reprogrammed into neural stem cells that tested younger than 20 on molecular aging clocks.
researchers in bonn basically managed to reset more than 60 years of epigenetic aging in human cells. the rejuvenation unfolded gradually over 100+ days, which means they can actually study the process and see what makes it happen.
the fact that we’re starting to understand how to reset the biological age of human cells is so fucking fascinating to me.
AI retweeted
Introducing Ling-3.1-flash-1019 🚀
It scores 44 on the Artificial Analysis Intelligence Index, up from 41 for Ling-3.1-flash. Gains in agentic work: AA-Briefcase ~1400→1524 Elo and AutomationBench-AA 62%→68%.
API access and open weights coming soon. Stay tuned.
AI retweeted
Scientists have discovered that COVID-19 can directly infect key brain cells responsible for dopamine production, driving them into an aged, 'zombie-like' state.
Researchers have uncovered a striking biological mechanism explaining why many individuals experience persistent brain fog, chronic fatigue, and loss of motivation following a COVID-19 infection.
Studies demonstrate that SARS-CoV-2 can directly target midbrain dopamine neurons—the specialized cells responsible for regulating movement, focus, and reward pathways. Upon infection, these neurons stop functioning normally and enter a state known as cellular senescence. Rather than dying, the infected dopamine cells become non-functional 'zombie' cells, halting dopamine production while actively secreting chemical signals that trigger ongoing localized inflammation.
This unexpected transformation mirrors biological pathways commonly seen in age-related neurodegenerative conditions like Parkinson's disease. By proving that dopamine neurons are uniquely susceptible to the virus, the findings provide a tangible target for medical interventions. Encouragingly, laboratory screens revealed that certain FDA-approved medications can prevent infection and rescue these damaged cells, opening promising new avenues for treating lingering neurological symptoms in long COVID patients.
source: Yang, L., Kim, T. W., Han, Y., Nair, M. S., Harschnitz, O., Zhu, J., Wang, P., Koo, S. Y., Lacko, L. A., ... Chen, S. SARS-CoV-2 infection causes dopaminergic neuron senescence. Cell Stem Cell, 31(2), 196-211.
AI retweeted
JUST IN: OpenAI launches global cybersecurity initiative called “Patch the Planet” after AI helped uncover critical flaws in PHP, a programming language powering millions of websites.
AI retweeted
Effective Immediately, all organizations using PHP for their systems are advised to be prepared for an update addressing two critical vulnerabilities. PHP is widely used for web applications. These vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it is recommend that organizations take the system offline until it can be fully mitigated. NSA will follow up shortly with more specific guidance.
By circa 2035 I envisage advanced AI-enabled robots scaling across manufacturing, retail (warehousing), military and security.
AI retweeted
Aloha!🌺 Meet Ornith-1.5-122B: No Tradeoff Between Strong and Fast
Stronger. Faster. Leaner. Ornith-1.5-122B delivers frontier-level capabilities with exceptional inference efficiency. With superior instruction following, stronger task decomposition, and deeper understanding of real-world workflows, it achieves state-of-the-art performance among open-source models of comparable size, while running on a single NVIDIA DGX card!!
Open weights under the MIT license.
Coding
✅SWE-bench Verified: 0.845
✅TB2.1: 0.813
✅CyberGym: 0.800
✅NL2Repo: 0.512
Agentic
✅HLE: 0.446
✅BrowseComp: 0.785
✅MCP-Atlas: 0.746
✅Toolathlon: 0.708
🤗Hugging Face: huggingface.co/ornith-ai/Orn…
We can’t wait to see what you build with Ornith-1.5-122B.
