Aabhash (@aa_bash_profile) header
Aabhash (@aa_bash_profile) profile picture

PhD Candidate: Decision-Focused ML | Optimization @ Uni Würzburg Interests: Energy/Transport. Prev. @Fraunhofer @BMW. Alum @UniFreiburg Tweets are personal.

Joined April 2010
Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)Photo by Aabhash (@aa_bash_profile)
Aabhash retweeted
Scarcity can only be allocated three ways: price, queue, or lottery. Pick your poison.
The FAA literally grounded every flight in the US over a nationwide communications outage, and Uber’s immediate response is to charge stranded passengers $100–$200 just to go a few miles from the airport. Profiteering off a national travel crisis isn't "dynamic pricing." It's predatory.
141
515
43
7,026
237,902
Should Europe leave frontier AI to the Americans and the Chinese? The idea has gained ground in Europe. It is a fundamental mistake. Build or surrender: for now, the choice is still ours. Full article below.
270
775
232
3,891
452,152
How AI shifted from a transductive paradigm (token-by-token answer completion) to an inductive paradigm (on-the-fly reasoning chain synthesis), in one chart. This is the big story of 2025 and 2026. This is what made coding agents work.
159
176
36
1,663
123,430
Singapore PM why his government doesn't cap prices to reduce cost of living pressures. What western leader today could explain the value of price signals in such a way?
85
683
119
5,263
854,656
Good point on why the layperson doesn't fully grasp or care about alignment yet. Posttraining/RL has allowed LLMs to deviate from the human-controlled space. Even further, moving from RHLF to verifiable rewards is another jump that the public doesn't grasp yet.
If I could add one thing to what the nontechnical world knows about AI, it would be what RL is and how it differs from pretraining/SFT. The general public has a solid grasp on pretraining (“predict the next token”), which at one point allowed many people to reason accurately about how LLMs work. But common knowledge has stagnated here. I think a similarly high-level understanding of RL makes it easy to see why we should expect both misalignment (RL instills alien traits, while SFT instills human ones) and extreme superhuman capabilities (RL doesn’t limit you to learning from stuff scraped from the internet). If I were not into AI and just knew the next-token predictor thing, I think I would be really confused as to why people are worried about this stuff, and then if you told me about RL I would become pretty easy to convince. I’m wondering what it would take to propagate an RL analog of “predict the next token” into the common intuition — I feel like this could be pretty big for generating political will, eg by turning people who think AI is a sham into people who are worried about real problems
2
178
रास्वपाका नेता-कार्यकर्ता, समर्थक-शुभेच्छुक प्रधानमन्त्रीको सतप्रतिशत गलत अभिव्यक्तिलाई ठीक साबित गर्न यो हद सम्मको हर्कत नगरिदिनुस्।एकदिन मै यसको असर हाम्रो अखण्डतामा कति परि सक्यो, मनन् गर्नुस्। देश भन्दा ठूलो अर्को चीज हुदैन। गलत भन्न सक्नुहुन्न भने मौन वसिदिनुस् । Please!
92
319
23
970
53,050
Such a great pleasure to witness the signing of an MoU between #TribhuvanUniversity and @Airbus. Nepal has exceptional young talent and this MoU is a concrete step towards preparing them to shape the future of aerospace in Nepal and beyond. -VC 28.05 PR: swll.to/6LopCvj
3
11
5
39
6,324
The hysteria over data centers is similar to the hysteria over nuclear power in the 70s and 80s. Our idiot politicians listened to morons who didn't understand that nuclear power is safe and clean. Now our power grids are aging and falling apart. More data centers. Full speed.
“wE dOn’T wAnT dAtA cEnTeRs!” 🤡 Guys, you can’t uninvent something. AI is here. It isn’t going away. If we don’t lead the world in AI, another country will. What does America look like when China or Iran or any other country supersedes us, and their tech runs circles around ours? People who want to slow down AI believe they’re doing good, but they aren’t. You think you’re saving jobs or keeping energy costs down or whatever altruistic ideas you have that motivate you. They are all wrong. In reality, they’re just shifting the power balance to countries that will never play by your rules and giving them the upper hand. We must lead the world on this. It’s non-negotiable.
260
258
50
2,407
177,097
The notion that developing countries can bypass manufacturing — and leap straight to services instead — ignores how every major economy actually developed.
52
669
64
2,308
85,600
This is the the quote I've been citing a lot recently.
you can outsource your thinking but you cannot outsource your understanding
847
4,354
272
46,575
2,784,460
The world is about to hit 612 Zettabytes of data by 2030. A standard laptop has 1 TB storage. So this is the equivalent of 612 billion laptops. Where is the next 500ZB of data going to live? We're trying to force-fit data centers into Virginia, London, Malaysia where the grid is already screaming. The physical internet is running out of space. Here is my analysis. 1/5
1
3
5
460
Cheap and clean means nothing if the country barely powers its own people. Building a massive data center in a country that's already short on power isn't ideal: blackouts, rationing, and energy bills that will crush households. Every server rack is a megawatt taken from someone’s house. So, in this map, I show countries that are projected to have surplus energy by 2030
1
1
4
213
With all of these points considered, where in the world are the best places to build new data-centers? 1. Cheap Electricity 2. Clean Generation 3. Surplus Energy (so we aren't pricing out locals) Four countries stand out. Paraguay 🇵🇾 Bhutan 🇧🇹 Nepal 🇳🇵 Ethiopia 🇪🇹 These four are the only ones that convincingly hit the "Cheap, Clean, Surplus" trifecta. The power is there. Now they just need the fiber.
5
195
Nepal will never be Switzerland. But if data is the new gold, a mountainous, neutral country should learn from 20th-century Swiss banking and start building 21st-century vaults. aabhash.de/articles/nepal-da…
2
6
268
Right now you’re largely experiencing the economy from a time when the Strait of Hormuz was open, because it takes many weeks for oil tankers to traverse the seas. You still experience the Strait as open, even though it’s been closed for weeks. But once that reality hits…
Paradox: If you see a baby located 90 light-years from Earth, right now it would be a 90-year-old, but you see it in your present, as a baby. While the light takes time to reach you, the baby grows and ages. When you look at the universe, you are always looking at the past.
128
4,529
91
36,870
1,141,832
This is the way forward for Nepal. Cheap, green electricity powering Data Centers and AI. Data centers need water based cooling as air cooling is expensive. Does not need a huge pool of human resource. Aligned with our geography, interests, resources. But need to move fast.
Amazing AI models are coming out of India. Yet the cost of running them is going to be huge. But I believe India can lead the world in locally run AI and give the cheapest AI access to our people - exactly like we did with cheap data that unlocked so much growth... ...By tapping the Himalayas as the base for AI clusters. Here's a simple idea I'm proposing for an AI data center: AI centers up in the Himalayas, may be in secure & stabilized tunnels, naturally cooled using pipes of ice cold air and frigid river water, powered by hydroelectric downstream. The entire Himalayan belt from Kashmir, Ladhak (potential for solar too), Uttarakhand, UP, Bihar can be used for these AI centers. Of course… this is just a rough idea to think about and try take it forward. There are many practical challenges. But I believe India can overcome them and lead the world in great but cheap and sustainable AI. Or I got a hunch China will definitely do it in the Himalayas.
4
253
Aabhash retweeted
Today we're introducing TRIBE v2 (Trimodal Brain Encoder), a foundation model trained to predict how the human brain responds to almost any sight or sound. Building on our Algonauts 2025 award-winning architecture, TRIBE v2 draws on 500+ hours of fMRI recordings from 700+ people to create a digital twin of neural activity and enable zero-shot predictions for new subjects, languages, and tasks. Try the demo and learn more here: go.meta.me/tribe2
718
2,433
1,486
15,864
6,985,287
ओलीको नेपाली राजनीतिमा उपादेयता लगभग सकिएको थियो। उनलाई कसरी फाल्ने भनेर एमाले सोच्दै थिए। गृहमन्त्रीले पक्राउ गरिदिए, उनि ५ दिन पनि हिरासत बस्दैनन्, तर अब ५ बर्षको लागि एमाले अध्यक्ष र सरकारको प्रतिपक्षी पक्का भए। त्यसैले त भनेको "राजनीतिक सुझबुझ भएको गृहमन्त्री चाहिन्छ"
1
4
1
23
2,448