Prashant Mathur

Last quarter I rolled out Microsoft Copilot to 4,000 employees. $30 per seat per month. $1.4 million annually. I called it "digital transformation." The board loved that phrase. They approved it in eleven minutes. No one asked what it would actually do. Including me. I told everyone it would "10x productivity." That's not a real number. But it sounds like one. HR asked how we'd measure the 10x. I said we'd "leverage analytics dashboards." They stopped asking. Three months later I checked the usage reports. 47 people had opened it. 12 had used it more than once. One of them was me. I used it to summarize an email I could have read in 30 seconds. It took 45 seconds. Plus the time it took to fix the hallucinations. But I called it a "pilot success." Success means the pilot didn't visibly fail. The CFO asked about ROI. I showed him a graph. The graph went up and to the right. It measured "AI enablement." I made that metric up. He nodded approvingly. We're "AI-enabled" now. I don't know what that means. But it's in our investor deck. A senior developer asked why we didn't use Claude or ChatGPT. I said we needed "enterprise-grade security." He asked what that meant. I said "compliance." He asked which compliance. I said "all of them." He looked skeptical. I scheduled him for a "career development conversation." He stopped asking questions. Microsoft sent a case study team. They wanted to feature us as a success story. I told them we "saved 40,000 hours." I calculated that number by multiplying employees by a number I made up. They didn't verify it. They never do. Now we're on Microsoft's website. "Global enterprise achieves 40,000 hours of productivity gains with Copilot." The CEO shared it on LinkedIn. He got 3,000 likes. He's never used Copilot. None of the executives have. We have an exemption. "Strategic focus requires minimal digital distraction." I wrote that policy. The licenses renew next month. I'm requesting an expansion. 5,000 more seats. We haven't used the first 4,000. But this time we'll "drive adoption." Adoption means mandatory training. Training means a 45-minute webinar no one watches. But completion will be tracked. Completion is a metric. Metrics go in dashboards. Dashboards go in board presentations. Board presentations get me promoted. I'll be SVP by Q3. I still don't know what Copilot does. But I know what it's for. It's for showing we're "investing in AI." Investment means spending. Spending means commitment. Commitment means we're serious about the future. The future is whatever I say it is. As long as the graph goes up and to the right.

Sad to hear about Pushpak sir! Rest in peace!

@elmanmansimov this reminds me of Gino's joke! youtube.com/watch?v=A-RfHC…

My 6 month old is picking up our commands to Alexa and now she's trying to say the wake word.. she's super fascinated by it @alexa99

Another great publication update - our paper comparing serialization strategies for entities in graphs has been accepted to NAACL! 🎉 Paper link will be updated soon!

Excited to share that our team has had two papers accepted to ICASSP this year! 🎉 - Improving lip-synchrony in audio-visual speech translation arxiv.org/abs/2412.16530 - Zero-resource speech translation arxiv.org/abs/2412.18566

Language Technologies Research Centre (LTRC) is celebrating 25 years of language research. Details at ltrc.iiit.ac.in/silver-jubilee/

If you're at #ECCV2024 do visit our poster (PS 4 session) at 10:30 AM local time tomorrow. Neither me nor @ilucasgoncalves could travel to the conference but amazing @kkundu10 will be presenting our work! @eccvconf

RLHF is a popular method. It makes your human eval score better and Elo rating 🚀🚀. But really❓Your model might be “cheating” you! 😈😈 We show that LLMs can learn to mislead human evaluators via RLHF. 🧵below

Snoo @happiestbaby is not meant for NYC babies. The bassinet cannot distinguish ambulance siren vs a crying baby. Any passing by ambulance causes Snoo to increase level of motion! 🤦‍♂️

@eccvconf I still haven't received the email yet and I am the corresponding author. Paper ID 10198. Our paper status is "Accept (dataset release)", does that affect these emails in any way?

@FaeghehSardari Emails have now been sent.

The #ECCV2024 camera-ready submission instructions for main conference papers has been sent to all authors via email. The deadline for submitting the camera-ready paper is **July 15 (22:00 CEST)**. By the same deadline you also need to cover the paper with a registration.

Today, Amazon announced its Trusted AI Challenge, which involves university students competing to securely advance large language models (LLMs) that code, and where $700,000 in cash prizes will be allocated across the top four performing teams. Learn more about the challenge and how to apply: amazon.science/trusted-ai-cha… #ResponsibleAI #GenAI

Our paper on Audio-Visual Synchrony Evaluation grounded in human perception is accepted at **ECCV 2024** !! Congrats to the amazing @ilucasgoncalves and other authors! arxiv.org/abs/2404.07336 Code and data is also released and useful to evaluate new metrics in this space!

What do World Cup victories mean to us? I don’t mean the players; we all know the answer to that. But what does it mean to us, the fans, the ones who do nothing? I can tell you what it means to me. They say that before one dies, the moments that mattered in our life flash before our eyes. World Cup victories will be in my final kaleidoscope. I will see Kapil Dev running back on the green of Lords, my little boy’s heart in my mouth, taking a catch that changes the history of a nation. I will see Misbah, with the Cup in hand, hand it over to the gentle dollies of Joginder Sharma, and I will see Dhoni, of course, sending the ball into the night sky. To that now, add Suryakumar Yadav, balancing himself on the line between victory and defeat, pivoting the game on a knife’s edge back to India. I will see them all, and I will see more. I will see who I was with when those moments came to pass. My father aging from 83 to 2024 in the images in my mind, and a new entrant to those memories, my daughter in 2024. I will see myself, too, reflected in the echoes of how I felt, from the boy wide-eyed in wonder in 83, hoping the power doesn’t go out again during the telecast to the young man in another country in 07 and 11 and now back in my own, this time a father, four different me-s in shape, size and mind, tied together through time simply by a chain of indescribable happiness, the happiness made happier by how rare these moments of World Cup victory are, dots on a timeline plagued by defeat and disappointment. But why this happiness for the achievements of a group of strangers, the cynical older me asks, one who play for a club called BCCI? Why bother? It’s because it’s these moments that tie our happiness to billions of others. It’s when the trajectories of our lives all meet at a common emotion at the exact same time before they diverge again in random ways, making us aware, even if for a fleeting moment, of a simple yet forgotten truth. We are not alone.

Congratulations @elmanmansimov !!!

@shiben_b I would've been more impressed if they did multimodal generation with audio+visuals combined. This is still the best Text2Video generated outputs I've seen so far!

Am I the only one who is meh 🫤 to #OpenAI #SoraAI ? Not because they are not impressive results, but because it's more of the same in the same content-gen space. What would make me go WOW 😲 when one product manager is able to build an entire app with zero engineers using AI

this is crazy good quality of videos!

@alexandraxron @Google Congratulations!

📢 Life update 📢 I'm beyond excited to share that I've joined @Google Bard as a Research Scientist! 🥳

📣 📣 📣 New instruction-tuned LLM! 📣 📣 📣 Today, we announce an initial release of "Airavata", an instruction-tuned LLM for Hindi. Blog: ai4bharat.github.io/airavata/ Model: huggingface.co/ai4bharat/Aira… Datasets: huggingface.co/datasets/ai4bh… (1/N)

@accezz nice! results look great as well!

We are releasing: * TowerBase, a continued pre-trained LLaMA2 * TowerInstruct, a finetuned TowerBase on a curated instruction set * TowerBlocks, the instruction set used for TowerInstruct huggingface.co/collections/Un…

Super happy to share something we have been working on lately: TowerLLM, a multilingual model geared towards cross-lingual and translation-related tasks. It has very good performance on translation benchmarks and supports 10 languages.