Joined July 2007
6,688 Photos and videos
Pinned Tweet
I've a paper out in @IPPRJournal on content blocking and Internet shutdowns in #India from the perspective of international human rights law. Comments welcome. Do share widely! PDF: ippr.in/index.php/ippr/artic… HTML: pranesh.in/scholarly/article… #FoE #censorship #IHRL
2
19
39
35,322
In some benchmarks, the city of Rio's post-trained Qwen 3.5 beats the Alibaba-released Qwen 3.7!
SITUATION DETECTED: The city of Rio de Janerio has post-trained a model. Based on Qwen 7/2, Rio 3.5 Open 397B adds SwiReasoning on top of the base Qwen model — a framework that dynamically switches between standard chain-of-thought and latent-space reasoning, guided by entropy-based confidence signals, so the model only "thinks out loud" when it needs to and otherwise reasons silently in hidden space for better token efficiency.
45
A mixture of models can perform better than any single model in that mix, by itself, can. #llm #genAI #AI #FOSS #OpenModels
Introducing the Fusion API, the smartest compound model in the market. Fusion achieves Fable-level intelligence at half the price. How it works 👇
78
Let's be clear: when Indians speak of "having our own foundation models", they aren't speaking of post-training excellent models to be even more suitable for Indians' needs; they're speaking of govt subsidies in creating foundation models from scratch.
The city of Rio has post-trained Qwen 3.5 397B to be an even better model foundation model by incorporating SwiReasoning (claiming Pareto superiority!): huggingface.co/prefeitura-ri… #DigitalSovereignty #OpenModels #AI #genAI #FOSS
1
1
206
All of India's ancient past was locked up because of a single script? Lols, no. The contributions of the British to Indian epigraphy, philology, etc., were enormous, but so was their exploitation of India. And what would they have studied were it not for Indian scribes, etc.?
Seriously if I were British, I’d make a page or YouTube channel just for this. Amelia teaching the world about British contributions. Real reasons to be proud of your ancestry and history. Never be ashamed of being great.
161
Rishad Premji of Wipro being eminently sensible. India (and most other countries) shouldn't get caught up in keeping up with the Joneses in terms of foundation LLMs: there's much more value to be unlocked in smaller models and in addressing actual use cases.
Replying to @shrutammegopaya
The leadership at Wipro OTOH seems convinced that the real opening for India is in "Small Language Models for specific applications", while giving cursory ceremonial lip-service to "multilingual AI". 🤷‍♂️ ndtv.com/video/india-s-ai-pl….
2
5
1,437
Pranesh Prakash retweeted
Joe Biden accidently triggered a movement to keep encryption free and open. newsletter.pessimistsarchive…
2
4
10
3,685
Pranesh Prakash retweeted
Notably, the budget panel was comparable with Claude Fable 5 in performance. A panel of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro, fused together, beat solo GPT-5.5 and solo Opus 4.8 outright. And it landed within 1% of Fable 5 while costing roughly half the price.
9
48
822
259,514
"Planning with Claude Fable 5 and implementing with GPT-5.5 produced the same service for 59% less than using Claude Fable 5 for both phases."
5
365
The city of Rio has post-trained Qwen 3.5 397B to be an even better model foundation model by incorporating SwiReasoning (claiming Pareto superiority!): huggingface.co/prefeitura-ri… #DigitalSovereignty #OpenModels #AI #genAI #FOSS
Alibaba Qwen3.7 slowly fading into irrelevance at the frontier due to proprietary stance. In it's place we have Minimax M3 and... *checks notes* Rio 3.5 397b, made by the municipal IT company of Rio de Janeiro's city government. huggingface.co/prefeitura-ri…
1
7
936
Pranesh Prakash retweeted
Breaking down US statement, that marks an escalation over the issue of 3 Indian seafarers being killed by US Navy, here's what it says: 1. Vessels must comply with US orders (Why?) 2. US forces are seeking to uphold peace and security( By shooting unarmed ships?) 3. In the Strait (But US ships are in the Gulf of Oman) 4. Violations of US Blockade ( which have been many) 5. Illicit transport of Iranian oil (Not according to international law, and US sanctions are only punitive economically) 6. Won't be tolerated (Regardless of loss of lives, or contravening international law) Awaiting MEA response. state.gov/releases/office-of…
US State dept readout of Jaishankar-Rubio call indicates sharp differences: "Secretary Rubio stressed that all commercial vessels should immediately comply with orders from U.S. forces as they seek to uphold peace and security in the Strait.  He underscored that violations of the U.S. blockade and the illicit transport of Iranian oil will not be tolerated.
113
840
2,161
145,837
Targetted therapies for cancer are making great strides. The great health policy question is: how do we make sure that these drugs actually reach people without waiting for 19–26 years for patent data exclusivity to run out.
Jun 12
All of these were reported over the past month: • A new pancreatic cancer drug, daraxonrasib, that roughly doubles survival in late-stage disease • A precision lung cancer drug, lorlatinib, that kept 55 percent of patients progression-free after 7 years, versus 3 percent on the old drug • A prostate cancer drug, talazoparib, that halves the risk of progression • An endometrial cancer drug, dostarlimab, where 58 percent of patients hadn't progressed after 4 years, versus 16 percent on chemo alone • An early-detection blood test, the NHS Galleri test, that quadrupled cancer detection but missed its main goal • An mRNA cancer vaccine that halved the risk of melanoma recurrence when added to Keytruda • The most effective weight loss drug so far, retatrutide, which cut body weight by about 28 percent • The first in vivo gene editing therapy, which cut hereditary angioedema attacks by 87 percent from a single injection • A one-time gene edit, VERVE-102, that lowered LDL cholesterol by 62 percent • A feat of pharmaceutical synthesis that raised enlicitide's manufacturing yield 14-fold using engineered enzymes • A functional cure for hepatitis B, bepirovirsen, that cleared the virus in about 20 percent of patients • The discovery that human cells can swap chromosome-sized DNA through nanotubes • An ancestor of CRISPR, VIPR, found in bacteriophages, that silences genes without cutting DNA • A preventive Covid-19 pill, ensitrelvir, that cut symptom risk by 67 percent after exposure • The first PROTAC drug, vepdegestrant, which destroys a disease-causing protein rather than blocking it Every month, Niko and I write a round up digging into the latest news in biotech and medicine, and this month's was astonishing. We share some thoughts on what's responsible for this progress and what it means for science in the future.
130
Pranesh Prakash retweeted
Can someone scan all the opinion pieces from the last one year to see how many recommended a “special mission” or a unit under the PMO for <issue of choice>? Policy (discourse and implementation) needs to expand beyond “make an exception for me” to “make things work normally”.
1
3
136
Pranesh Prakash retweeted
No words needed 💔
298
3,096
9,280
97,567
Pranesh Prakash retweeted
GLM-5.2 is Fully Open, Frontier Intelligence Belongs to Everyone Today, the sudden restriction of certain frontier models is deeply regrettable. At a time when access to frontier models is abruptly cut off for non-technical reasons, we are even more convinced of one thing: science should be global. The path to AGI (Artificial General Intelligence) must never be enclosed by high walls. We have always believed that AGI should be the cornerstone for all of humanity to collaboratively explore the boundaries of intelligence and solve complex challenges, rather than a privilege monopolized by a few rules and subject to revocation at any moment. In the face of external blockades and restrictions, our attitude is one of radical openness. Frontier intelligence must remain open-source, accessible, and buildable, serving every dedicated developer. GLM-5.2 is Zhipu's most capable open-source model to date. It not only supports a truly usable 1M context window but also maintains a continuous lead in the independent completion of long-horizon tasks, providing solid foundational support for building complex agent applications. It also continues to be our main engine for creating the strongest domestic coding model. Tonight at 5:21—at this special moment—GLM-5.2 will officially be available to all GLM Coding Plan users (including Lite / Pro / Max). The API will also go live next week. A step closer to frontier intelligence for everyone. The future of AI is open, and it is for the people. ModelKey: GLM-5.2
204
558
5,710
634,302
Pranesh Prakash retweeted
"it is not a conflict between US and Chinese tech; it is a battle between dependency versus digital sovereignty" foss AI models from any country >>> indian (or any other) proprietary models
In light of the news of the US export licence on Anthropic's Fable/Mythos, I'm re-upping my op-ed, in which I argued that India must seek digital sovereignty in the form of open AI models because access to proprietary models can be revoked at any time. pranesh.in/press/mythos-self…
1
207
Pranesh Prakash retweeted
Are we putting Deepseek v4 Pro, Kimi K2.7-Code, MiniMax M3, GLM 5.1, Gemma, and other very capable open source AI models to optimal use? The answer is, "No." We shouldn't fall prey to keeping up with the Joneses while underutilizating existing resources. #DigitalSovereignty
1
1
219
Pranesh Prakash retweeted
Some reads from the current Fable ban situation: - Vagueposting that a model can hack everything has consequences if you then end up releasing anyway. Saying other models also can after the fact is not enough. - Asking for regulation when you can't specify exactly what regulation has predictable consequences. - The ratchet is clearly moving towards license raj - There are many who want an implicit license raj (AISI testing with power to block) but it's the same thing in practice. It is bad. Bad for safety, since now there's no choice but to accelerate for others. - There's no way to allow models to be used "at large" going forward if the govt treats models as weapons. - This is *fantastic* for Chinese models. - The govt is ofc overstepping but honestly if you didn't expect that then you're naive! - Leopold's narrative is almost too on the nose. - Safetyists have wanted "perfect safety" as a goal, which is unachievable, and I've said a thousand times before it will backfire. This is the backfire. - This *still* assumes the old view that the individual model is the bad part and not a system, which will inevitably lead to bad governance. - This will get reversed in a bit and the model will get released (license raj), but the precedent is set. And many will say "ah this was bad but at least we got a license raj". They will be wrong. - Openai has more breathing room for a better model to be released. And they're toning down the rhetoric. This will help them. - Competition is good.
16
47
339
31,097
Pranesh Prakash retweeted
Perfectly strange until Joe Sacco revealed in his conversation with The Wire’s @bombaywallah that it wasn’t just one map but five pages of things that the publishers wanted changed. Sacco said he “got fed up” and refused to make any changes. @seemay✍️ thewire.in/books/future-riot…
2
22
76
11,205