Mitul Tiwari

🔍 Check out my blog post detailing DeepSeek R1's new reinforcement learning technique Group Relative Policy Optimization (GRPO) - eliminates value models while maintaining PPO's stability, potentially slashing training costs: linkedin.com/pulse/deepseek… #AI #MachineLearning #LLM #RL

Excited to announce that TapeAgent-the AI agent framework-is now open-source! 🎉 📼 Records all agent actions, thoughts, and observations 🔍 Simplifies debugging, replay, and analysis 📈 Boosts AI performance with data-driven optimization Learn more: linkedin.com/pulse/introduc…

In recent times, the AI community has been excited over the emergence of a mixture of expert llms, such as Mixtral. This blog post dives deeper into the architecture of these innovative models, exploring their potential and pointers for further reading. linkedin.com/pulse/mixture-…

Thrilled to introduce RefuelLLM-2, our latest family of LLMs built for data labeling and enrichment tasks. RefuelLLM-2 (83.82%) outperforms GPT-4-Turbo (80.88%), Claude-3-Opus (79.19%), Llama3-70B (78.2%) and Gemini-1.5-Pro (74.59%) on a benchmark of ~30 data labeling tasks: RefuelLLM-2-small (79.67%), aka Llama-3-Refueled, outperforms all comparable LLMs including Claude3-Sonnet (70.99%), Haiku (69.23%) and GPT-3.5-Turbo (68.13%). We’re open sourcing the model: huggingface.co/refuelai/Llama… You can try out the models here and give us some feedback! labs.refuel.ai/playground. The code and data used for benchmarking the LLMs is available in our Autolabel library: github.com/refuel-ai/auto… One more thing: RefuelLLM-2 family of models output much better calibrated confidence scores - a useful lever to reject, retry or ensemble low confidence outputs.

We at @FoundationCap believe there is $4.6T of work to be automated. AI companies are leading a transition from Software-as-a-Service to Service-as-Software, turning the table on the very essence of SaaS. We look at the areas to be automated in two buckets: 1.) Salaries of jobs globally ($2.3 trillion in sales & marketing, software engineering, security, and HR) 2.) The amount spent on outsourced services and salaries—both IT services and business process services ($2.3 trillion, per Gartner) In the software business, a company may sell access to its platform or tool, but customers are still responsible for using that tool to achieve the desired outcome. In the services business, responsibility for achieving the desired outcome sits with the company selling the service.

Check out my new blog on Domain Adaptation and Aligning to Human Preferences of LLMs like Mistral & Llama2! This post gives an overview of domain adaptation techniques, and dives deeper into alignment with direct preference optimization (DPO): linkedin.com/pulse/domain-a… #AI #LLM

Large Language Models II: Are you fascinated by the advancements in LLMs, and curious about Transformer, Self-Attention, and the architecture of SOTA LLMs like Mixtral & Llama-2. If yes, then you can't miss the 2nd installment of this series on LLMs: linkedin.com/pulse/large-la… #LLM

I attended BayLearn 2023 conference and presented our work on "Zero and Few-shot Techniques for Intent Classification using LLMs". In this post I cover highlights from the conference. linkedin.com/pulse/thoughts… #baylearn #ai

Using GPT-3 for Data Augmentation to Recognize Dialog Act. A summary of our recently published EMNLP GEM paper linkedin.com/pulse/using-gp…

Language models have changed the world of NLP in the last 10 years. This article covers one of the first language models that came out in 2013. Check it out and stay tuned for the subsequent posts on more recent language models. #nlp, #conversationalai linkedin.com/pulse/language…

@parul8ue @Springboard Congratulations @parul8ue on the accomplishments. All the best for future.

1/ It's the end of an era. After 8 wonderful years, I am taking off my operator hat @Springboard. Will continue on as a board member & big cheerleader

Interesting.

@danbrettig @ESPNcricinfo @ESPNcricinfo @danbrettig did someone ask Smith why was he erasing Pant's guard marks? There is no explanation! #guardgate cc: @ICC @CricketAus @BCCI

Paine was adamant that Smith had meant nothing sinister or annoying to Pant by what he had done on the crease line. Even so, Paine conceded that Smith would need to reconsider venturing into those areas in future "because of the perception of it" #AUSvIND espncricinfo.com/story/australi…

Nice to see Passage AI acquisition by ServiceNow as one of the top technology M&A deals of 2020! cc: @ravi_raj @mathihalli lnkd.in/deJZN-E

Looking forward to review papers for TheWeb2021 (lnkd.in/d44m3C7) personalization track as a part of the program committee. Recently reviewed WSDM 2021 papers and got to read some good work on deep learning for conversational AI.

For the 7th year in a row, #Gartner names ServiceNow a Leader in the ITSM Tools Magic Quadrant. Get the full report: spr.ly/6014G2elg #ITSM

Looking for an experienced machine learning engineer to work on cutting edge deep learning technologies to build conversational AI and natural language understanding. Please reach out if you are or know someone who would be interes…lnkd.in/dVwpfp9 lnkd.in/difVZ4F

[New Post] The economics of AI / ML companies are poorly understood. In this post, @BornsteinMatt and I cover some of the core challenges and how they are being tackled by top teams in the industry. a16z.com/2020/08/12/tam… Highlights below 👇

13/ THANK YOU to @Aman_Naimat, @vitalygordon, @Hassan_Sawaf, @zaydenam, @alex_holub, @evanrsparks, @mitultiwari, @spiantino and others whose twitter handles we couldn’t figure out :) Full post : a16z.com/2020/08/12/tam…

Insightful talks by Grace Huang on personalized messaging system (datacouncil.ai/talks/mesa-bui…) and @mitultiwari on conversational AI agents (datacouncil.ai/talks/lessons-…). @DataCouncilAI will be posting the videos/slides online in the near future. Thanks @petesoder, Anna Heim & speakers!