Kartik Talamadupula
@kr_t
Followers
686
Following
687
Media
222
Statuses
2K
Head of AI @WandAI_ | Ex- @symbldotai @IBMResearch @ASU | AI Planning, NLP, ML, GenAI, Conversational AI | Startup Advisor | AAAI Senior Member | Views my own
Seattle, WA
Joined June 2010
Check out our latest work (that @rasbt talks about in much detail in the post below) on "Concise Reasoning via Reinforcement Learning" -. Paper: Blog:
As we all know by now, reasoning models often generate longer responses, which raises compute costs. Now, this new paper ( shows that this behavior comes from the RL training process, not from an actual need for long answers for better accuracy. The RL
0
0
1
RT @rao2z: ICYMI, THE great Ted Chiang gave a very thoughtful speech on GenAI and Art at the #NeurIPS2024 workshop on GenAI and Creativit….
0
43
0
RT @drjingjing2026: 1/3 Today, an anecdote shared by an invited speaker at #NeurIPS2024 left many Chinese scholars, myself included, feelin….
0
565
0
The best part of conferences - and #NeurIPS2024 is no exception - is catching up with colleagues, collaborators, friends: both old and new! . Here in these pictures, various faces from the past and the present, including advisors, mentors, managers, and first paper co-authors! ❤️
0
0
6
RT @rao2z: With some Yochanites--past and present--at #NeurIPS2024 🤗 (L-R: @kayastechly @karthikv792 @rao2z @sarath_ssreedh @kr_t ) https:/….
0
5
0
The competition at the poster sessions at #NeurIPS2024 is intense. How intense? . If your poster doesn't mention LLMs by the third t̶o̶k̶e̶n̶ sentence, this is where it ends up. 😞
1
0
2
Bravely holding the fort against all the "AGI" prognostications 💪.
With some Yochanites--past and present--at #NeurIPS2024 🤗 (L-R: @kayastechly @karthikv792 @rao2z @sarath_ssreedh @kr_t )
0
0
4
OpenAI (again) - . Look o2 can now predict how many "o" s in Doom. Also in P(Doom)!!1 🙃.
OpenAI comms: [underspecific hype-y 'big tings coming!!' pls like and subscribe].Google comms: [corporate vagueness about Gemini3-0011 v2 FINAL.docx on Vertex available to 14 users].GDM comms: [we have simulated a rat's brain capable of solving 4D chess, but we're not sure why].
0
0
2
When they said "we will see the first one-person billion dollar company soon", I think they forgot to specify the direction of causation. And that, kids, is why causality is important.
At the rate the exodus is going and the progress is expected to go, . by 2027 spring OAI will just be Sam and GPT9. By that summer, it will be just GPT10 on Three Mile Island. True capitalist utopia. 🤞🤗.
0
0
2
RT @rao2z: So @karthikv792 is hard at work transferring our research funds to @sama. er. evaluating o1 on the PlanBench. Will share when….
0
13
0
The YouTube livestream of the talk is available here: See you there!.
0
0
0
Excited to be giving an invited talk at the Neuro-Symbolic AI Summer School 2024 in 30 min! I will discuss if human conversations are special to LLMs, and if today's language models truly understand humans. Click here: for video stream.
1
1
3
@GeorgiaTech Resources:. Research paper: Blog explaining the paper: Dataset (Github): Dataset (@huggingface):
0
0
1
If you want to learn more, check out our research paper and dataset for a deeper dive into the methodology and findings (links follow in the next tweet). We have also created a blog post that provides a more .accessible explanation of our work. Joint work with @GeorgiaTech.
1
0
0
This work has broad implications for the future of AI assistants and conversational interfaces. Imagine AI and language models that can not only understand what you say but also *how* you say and mean it, leading to more human-like interactions.
1
0
0
Our results show significant improvement in sarcasm detection compared to traditional methods. By incorporating "Reading with Intent," LLMs (and entire LLM families) can better interpret and respond to sarcastic remarks, leading to more accurate and natural conversations.
1
0
0
We developed a novel dataset specifically designed to train and evaluate LLM families on sarcasm detection within a multimodal Retrieval Augmented Generation (RAG) framework. This dataset, featuring sarcasm and fact "poisoned" data, is publicly available now. (links below)
1
0
0
Sarcasm often relies on understanding subtle cues beyond just the words that are rendered. Our approach focuses on the intent of the words, analyzing both text and emotional cues to identify sarcastic utterances. This is crucial for building truly conversational AI.
1
0
0
Ever wondered how to make Large Language Models (LLMs) understand sarcasm, especially in multimodal contexts like video calls? . Our new research at @symbldotai explores just that! . We introduce the concept of "Reading with Intent" to equip LLMs with this nuanced capability. 🧵
2
1
3