Pan Lu
@lupantech
Followers
4,687
Following
1,065
Media
195
Statuses
808
Postdoc @Stanford | PhD @CS_UCLA @uclanlp | Amazon/Bloomberg/Qualcomm Fellows | Ex @Tsinghua_Uni @Microsoft @allen_ai | Math Reasoning, AI4Science, #NLP , LLMs
Palo Alto
Joined April 2016
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
America
• 1075191 Tweets
Happy 4th
• 875103 Tweets
Labour
• 610827 Tweets
Independence Day
• 523748 Tweets
Reform
• 488385 Tweets
Tories
• 290903 Tweets
#loveIsland
• 217427 Tweets
Tory
• 203913 Tweets
Keir Starmer
• 130811 Tweets
#GeneralElection2024
• 119826 Tweets
Mimi
• 109754 Tweets
Sean
• 103380 Tweets
Mario Delgado
• 74126 Tweets
Maya
• 61849 Tweets
Corbyn
• 54858 Tweets
#TemptationIsland
• 40370 Tweets
Sky News
• 32759 Tweets
Raul
• 32740 Tweets
Andy Murray
• 31305 Tweets
Channel 4
• 26233 Tweets
Luca
• 25227 Tweets
Lib Dems
• 23625 Tweets
Reino Unido
• 20009 Tweets
Matilda
• 18184 Tweets
THE ARCHER
• 15275 Tweets
#ExitPoll
• 12984 Tweets
Joey Chestnut
• 10571 Tweets
GUILTY AS SIN
• 10144 Tweets
Pinned Tweet
I'm excited to join Prof.
@james_y_zou
group as a postdoc scholar, aiming to push the boundaries of AI for scientific discovery
#AI4Science
.
I've had an incredible and rewarding time with the
@uclanlp
group and the VCLA group
@UCLAComSci
. Deeply grateful to all my mentors,
15
2
217
🔥Excited to release LLaMA-Adapter! With only 1.2M learnable parameters and 52K instruction data, LLaMA-Adapter turns a
#LLaMA
into an instruction-following model within ONE hour, delivering high-quality responses!
🚀Paper:
🚀Code:
24
174
820
🔥Thrilled to release LLaMa-Adapter Multimodal!
🎯Now supporting text, image, audio, and video inputs powered by
#ImageBind
. 🧵6
💻Codes for inference, pretraining, and finetuning ➕ checkpoints:
demo:
abs:
15
149
640
🎉Exciting news: LLaMA-Adapter is now fully unlocked! 🧵6
1⃣ As a general-purpose
#multimodal
foundation model, it integrates various inputs like images, audio, text, video, and 3D point clouds, while providing image, text-based, and detection outputs. It uniquely accepts the
22
166
603
🚨 BREAKING:
@OpenAI
's new GPT-4o model outperforms humans on MathVista for the first time!
📊 Scores:
Human avg: 60.3
GPT-4o: 63.8
📖 Learn more:
OpenAI :
MathVista:
We're opening up access to our new flagship model, GPT-4o, and features like browse, data analysis, and memory to everyone for free (with limits).
546
3K
15K
8
89
522
🚀Introducing
#LLaMA2
-Accessory - an advanced open-source toolkit for large language models.
Evolved from LLaMA-Adapter, we now support more datasets, tasks, visual encoders, and efficient optimization methods.🧠
🔗Code:
💡Key Features:
🎯 Pre-training
13
134
505
🚀65B LLaMA-Adapter-V2 code & checkpoint are NOW ready at !
🛠️Big update enhancing multimodality & chatbot.
🔥LLaMA-Adapter-V2 surpasses
#ChatGPT
in response quality (102%:100%) & beats
#Vicuna
in win-tie-lost (50:14).
☕️Thanks to Peng Gao &
@opengvlab
!
2/2
11
102
411
🎉New paper! The survey of deep learning for mathematical reasoning (
#DL4MATH
) is now available. We've seen tremendous growth in this community since 2018, and this review covers the tasks, datasets, and methods from the past decade.
Check it out now:
6
79
337
LLaMA-Adapter V2, the next-gen multi-modal instruction model, boasts a model size multiple times larger than 7B! 🌟🔥
Chatbot systems, get ready for a major upgrade! 🤖💬
Stay tuned! Technical report & models coming soon. 📄🔜Keep up to date!
🔗
4
62
312
🚀Excited to release our 112-page study on math reasoning in visual contexts via
#MathVista
. For the first time, we provide both quantitative and qualitative evaluations of
#GPT4V
,
#Bard
, & 10 other models.
📄✨Full paper:
🔗Proj:
16
79
313
Congrats,
@JeffDean
@GoogleDeepMind
! Gemini 1.5 Pro has shown substantial improvements from Feb to May, scoring 63.9% on our
#MathVista
(), outperforming humans and GPT-4o, which was out 4 days ago!🚀
AI Progress has never been this rapid and impressive!🌟
Gemini 1.5 Model Family: Technical Report updates now published
In the report we present the latest models of the Gemini family – Gemini 1.5 Pro and Gemini 1.5 Flash, two highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information
28
235
994
8
63
305
🚀 Introducing
#SPHINX
: The Next-Gen
#Multimodal_LLM
. Seamlessly blending Tasks, Embeddings & Weights for advanced multimodal reasoning. 🧵N
🔍Demo:
💻Code:
What's New with
#SPHINX
compared to
#LLaMA_Adapter
? 🆕
✅ Powered by the
12
67
274
🚀Meet Chameleon! An innovative plug-and-play framework enhancing
#GPT4
and
#ChatGPT
like
#AutoGPT
for compositional reasoning, blending off-the-shelf tools with tailored LLM models 🔧✨🧠. New SOTA on
#ScienceQA
and TabMWP! 📈
🔗
📜
14
73
260
🚀 Introducing the LLaMA-Adapter, now available on
@huggingface
!
🔗
🎉 Feel free to explore and experiment with our LLaMA-Adapter. We're eager to hear your feedback!
💥 Stay tuned for the upcoming second version - even more powerful and feature-packed!
3
41
246
🎉 Thrilled to have our MathVista work accepted at
#ICLR2024
as an Oral presentation!
Explore our work:
🔍 Project:
🤗
@huggingface
Dataset
@_akhaliq
:
💻 Code:
Deepest gratitude to our shining team: 👏🌟
🚀Excited to release our 112-page study on math reasoning in visual contexts via
#MathVista
. For the first time, we provide both quantitative and qualitative evaluations of
#GPT4V
,
#Bard
, & 10 other models.
📄✨Full paper:
🔗Proj:
16
79
313
7
33
247
I am thrilled to defend my PhD and finally earn the title of Doctor🧑🎓. It's been a truly rewarding journey at
@UCLAComSci
. I'm so fortunate and grateful for the invaluable mentorship from Prof.
@kaiwei_chang
@uclanlp
. He has always been incredibly encouraging, helpful, and
Congrats 🎉 to the newly titled Dr. Lu
@lupantech
on defending his thesis about mathematical reasoning with language models"! 🧮 Pan has published a series of works on quantifying and improving math and scientific reasoning ability in LLMs. Some highlights:
1
5
82
42
2
233
🔥Boost your GPT-3 with our ICLR-23 paper on PromptPG! The first of its kind, PromptPG uses RL to select optimal examples for GPT-3, leading to a 5.31% gain on the TabMWP dataset of math word problems. Don't miss out on this game-changing solution!
👉 🧵1/7
2
30
226
🔥 Introducing
#SPHINX
🦁: an all-in-one multimodal LLM with a unified interface that seamlessly integrates domains, tasks, & embeddings. 🧵N
👋 Explore the
@Gradio
demo
@_akhaliq
:
Dive into the open resources!
🤗 Model
@huggingface
:
13
52
211
🔍 Does Multi-modal LLMs Truly Understand Diagrams in Visual Math Problems?
🧐 Interest in visual math reasoning has surged in the era of Multi-modal LLMs (
#MLLMs
). Although showing promising potential, it remains uncertain whether MLLMs utilize visual or textual shortcuts to
MathVerse
Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?
The remarkable progress of Multi-modal Large Language Models (MLLMs) has garnered unparalleled attention, due to their superior performance in visual contexts. However, their capabilities in
1
72
259
1
34
211
🎉 Just reached 1000 citations on Google Scholar! Grateful to be part of a community that values and engages with my research. Here's to continued curiosity and exploration! 🔍
7
0
189
🤔 Ever wondered why foundation models like LLMs & LMMs are only tested on textual math reasoning benchmarks?
🔍 Dive into our
#MathVista
for a fresh perspective: !
🌟 Introducing
#MathVista
: A groundbreaking benchmark for visual mathematical reasoning –
13
49
186
🌟Last week, I am honored to present our latest work
#Chameleon
to the Reasoning Team at Google Brain
@DeepMind
. It's encouraging to witness tool-augmented LLMs like Transformer Agents
@huggingface
and Chameleon garnering significant attention. 🧵6
Slides:
4
33
165
Model editing has been an effective way to reduce hallucinations in LLMs, instead of undergoing resource-intensive retraining.
🤯However, our study, led by
@JasonForJoy
,
@kaiwei_chang
, &
@VioletNPeng
, reveals that current methods inadvertently impair the general skills of LLMs.
1
30
159
🚨Struggling to select examples for GPT-3? Try our PromptPG, the first work that applies RL to select in-context examples for GPT-3! PromptPG achieves a gain of 5.31% on TabMWP, a new dataset of tabular math word problems! Check out data and codes:👇
🧵1/7
2
22
154
🚨Thrilled to have one paper accepted to
#NeurIPS2022
! We construct a new benchmark, ScienceQA, and design language models to learn to generate lectures and explanations as the chain of thought to mimic the multi-hop reasoning process. Data and code will be coming soon!
2
14
147
📢📢Excited to have one paper accepted to
#NeurIPS2022
! We present a new dataset, ScienceQA, and develop large language models to learn to generate lectures and explanations as the chain of thought (CoT). Data and code are public now! Please check👇👇
4
27
145
🔥 Exciting Update! We've manually evaluated
#GPT4V
using the playground chatbot on
#MathVista
, our newest benchmark for visual mathematical reasoning.
🚀
#GPT4V
soared with a 15.1%⬆️ improvement over
#Bard
, setting a new record at 49.9%! 🎉
🌐
Yet,
3
28
135
Our
#Chameleon
ranked
#1
among 1682 AI papers last week by
@alphasignalai
, emphasizing the significant impact our work has made.
#Chameleon
is a plug-and-play reasoning framework, enabling LLMs to utilize diverse tools.
🔗
🎉 More:
1
35
131
🤖 Could
#LLMs
develop emotional intelligence to undestand human social interactions?
Introducing KokoMind 🦍: a benchmark to evaluate how
#gpt4
,
#chatgpt
, &
#claude
interpret conversations and relations, and contribute with insightful advices.
💥 Demo:
Put ChatGPT at a cocktail party🥂.
Can it
- understand people's conversations, gestures
- figure out their relations,
- and even chime in with social advice?
🦍Announce KokoMind.
🌟Check out this demo! More at
#AI
#GPT4
#ChatGPT
#OpenAI
#Shrinking
🧵
13
88
305
4
26
127
Thrilled to be awarded the prestigious
@Bloomberg
#DataScience
Ph.D. Fellowship! 🏆 Grateful for the support and mentorship from
@TechAtBloomberg
to advance my AI research, especially in LLMs.
Heartfelt thanks to
@kaiwei_chang
@uclanlp
&
@UCLAComSci
for their tremendous support!
Congratulations to
@UCLAComSci
/
@UCLAengineering
+
@uclanlp
's
@lupantech
on being one of the 2023-2024
@Bloomberg
#DataScience
Ph.D. Fellows!
Learn more about Pan’s research focus and our latest cohort of Ph.D. Fellows:
#AI
#ML
#NLProc
#LLMs
0
0
5
5
4
111
Introducing
#STIC
: A Self-Training Method for Large Vision Language Models (LVLMs)! 🌟 🧵
STIC empowers LVLMs to self-train and enhance reasoning abilities using self-constructed preference data on image descriptions, eliminating the need for labeled data! 🚀📈
Straightforward
7
20
102
🚀 Introducing MuirBench! 🌟
A groundbreaking benchmark for robust multi-image understanding, featuring:
📸 12 diverse tasks
🗂️ 10 categories of multi-image relations
🖼️ 11,264 images
❓ 2,600 multiple-choice questions
Even top models like GPT-4o and Gemini Pro find it
Can GPT-4o and Gemini-Pro handle 𝐦𝐮𝐥𝐭𝐢𝐩𝐥𝐞 𝐢𝐦𝐚𝐠𝐞𝐬?
Introducing MuirBench: A Comprehensive Benchmark for Robust Multi-image Understanding.
🌐 Explore here:
📄 Paper:
📊 Data:
2
42
94
2
14
100
🚀🎉 Introducing X-Accessory's new member: Large Diffusion Transformer (Large-DiT)! 🎆✨
🔗
💪 We're pushing boundaries by expanding diffusion transformers to 7B parameters. Here are our features: 🧵6
1⃣ Model Scaling-up 📈: Scale to 3B and 7B by merging
7
20
98
Can machines answer multi-modal math word problems? We proposed a new task, Icon Question Answering
#IconQA
, to deal with it!
Details are available below:
Paper:
Project:
Code:
3
25
96
Excited to meet
@ylecun
with the
@uclanlp
labmates
@JasonForJoy
,
@LiLiunian
, and
@ZiYiDou
! 😝
#NeurIPS2023
0
3
94
📢 Can't wait to see you at the 3rd
#MathAI
Workshop in the LLM Era at
#NeurIPS2023
!
⏰ 8:55am - 5:00pm, Friday, Dec 15
📍 Room 217-219
🔗
📽️
Exciting Lineup:
⭐️ Six insightful talks by
@KristinLauter
,
@BaraMoa
,
@noahdgoodman
,
4
21
88
🤖In sciences and finance, we often engage in statistical and causal reasoning with structured data. Ever dreamed of
#LLMs
doing the heavy lifting, clearing the path from the maze of complex and error-prone tasks? 🤯
Hold that thought! 🛑 Our findings reveal that even GPT-4
Are LLMs Capable of Data-based Statistical and Causal Reasoning?
In this work, we propose a benchmark QRData (Quantitative Reasoning with Data) to evaluate models' capability in statistical and causal reasoning with real-world data.
🌐:
1
24
81
0
21
88
I am honored to win the
@Qualcomm
Innovation Fellowship! A heartfelt thank you to
@kaiwei_chang
for your kind words and encouragement. I am grateful to our team, including
@liujc1998
and Professor
@HannaHajishirzi
. This achievement wouldn't have been possible without you all! ❤️
Congrats
@lupantech
for winning the 2023 Qualcomm Innovation Fellowship! 🐻 Pan is a rock star in math and scientific reasoning in NLP!
0
3
20
3
5
86
🔥Thrilled to announce that our LLaMA-Adapter has been featured in Lit-LLaMA by
@LightningAI
🦙🦙
🚀 Check out our LLaMA-Adapter here:
⚡️ Explore Lit-LLaMA on GitHub:
Progress update!🦙🔥🤓
Lit-LLaMA now implements the LLaMA-Adapter method for efficient fine-tuning 🔧⚡️
The core idea can be implemented in about 11 lines of code🤯 (see screenshot)
Link to repo👉
Link to Adapter paper👉
2
41
170
2
12
85
💥💥Update Alert! Radar graphs & leaderboard on
#MathVista
now feature detailed scores for the
#Gemini
family models. 🚀
🔍 Insight: Gemini Ultra leads the pack, outperforming GPT-4V by 3.1%! Yet, each model shines uniquely in various math reasoning & visual contexts.
🙏 Big
2
16
83
Privileged to have the opportunity to guest lecture on
#NLP
course
@CS_UCLA
, instructed by Prof.
@kaiwei_chang
. I really enjoyed it and am so glad to share recent advancements in mathematical reasoning and commonsense reasoning.🧵3
🔗Check out the slides:
4
7
79
Hey Friends! 🎉 Excited to be at
#NeurIPS2023
! 🚀 I’ll be presenting a paper 📄, co-organizing the MATH-AI workshop 🧮, and sharing three collaborative projects. Can't wait to meet you in New Orleans 🎭 and explore the AI advancements in math, science, and more! 🤖🧪
👇1⃣2⃣3⃣4⃣
1
5
78
🦙Please check out LLaMA-Adapter-V2, performing open-ended multi-modal visual instructions by merely introducing 14M learnable parameters over 65B
#LLaMA
.
abs:
repo:
weights:
video:
🚀65B LLaMA-Adapter-V2 code & checkpoint are NOW ready at !
🛠️Big update enhancing multimodality & chatbot.
🔥LLaMA-Adapter-V2 surpasses
#ChatGPT
in response quality (102%:100%) & beats
#Vicuna
in win-tie-lost (50:14).
☕️Thanks to Peng Gao &
@opengvlab
!
2/2
11
102
411
0
22
78
Excited to see the release of Gemini!
It is more excited to see that Gemini
@google
features MathVista for evaluating math reasoning in visual contexts and Geometry3K for evaluating geometry reasoning!!
Congratulations and thanks
@GoogleDeepMind
,
@GoogleResearch
, and
@Google
!
I’m very excited to share our work on Gemini today! Gemini is a family of multimodal models that demonstrate really strong capabilities across the image, audio, video, and text domains. Our most-capable model, Gemini Ultra, advances the state of the art in 30 of 32 benchmarks,
276
3K
13K
1
5
75
We're organizing the 3rd
#MathAI
workshop at
@NeurIPSConf
#NeurIPS
.
🚀 Excited for our speakers on AI for mathematical reasoning,
@guyvdb
,
@noahdgoodman
,
@wtgowers
,
@BaraMoa
,
@KristinLauter
,
@TaliaRinger
,
@paul_smolensky
, Armando Solar-Lezama,
@Yuhu_ai_
,
@ericxing
,
@denny_zhou
.
0
12
70
Today, we presented our
#MathVista
() at
#ICLR2024
in Vienna! 🌟
We are thrilled by the tremendous progress in math reasoning in the era of LLMs and VLMs. MathVista has become one of the most reliable benchmarks for probing their abilities in visual math
🚀Excited to release our 112-page study on math reasoning in visual contexts via
#MathVista
. For the first time, we provide both quantitative and qualitative evaluations of
#GPT4V
,
#Bard
, & 10 other models.
📄✨Full paper:
🔗Proj:
16
79
313
5
9
69
Spent a fantastic weekend at Lake Arrowhead with the
@uclanlp
group! ❄️🏔️⬆️ Enjoyed scenic drives, delicious meals, engaging conversations, and brainstorming sessions. Truly inspiring! 🚗🥘😋💬 🖼️🧠💡
2
6
68
📢Great news! Our
#ScienceQA
dataset is gaining significant attention lately. It is the primary benchmark for the next-gen
#MultimodalCoT
reasoning system by
@AmazonScience
, and it's now included in
@huggingface
: .
More details: 👉
1
15
67
It is my great honor to be awarded the
#Bloomberg
Data Science Ph.D. Fellowship! Many thanks to the tremendous support from
@TechAtBloomberg
,
@UCLAComSci
, and Professor
@kaiwei_chang
@uclanlp
! Go Bruins🐻✊!
CS Ph.D. Pan Lu Awarded Bloomberg Data Science Ph.D. Fellowship
Read more:
0
0
10
2
1
63
@_arohan_
@JeffDean
@GoogleDeepMind
Hi Rohan, thanks for pointing it out. We have updated the leaderboard with Flash. Congratulations to you and your team on the development of these impressive models! 🏆
3
7
59
🛠️🚀 Excited to share our latest paper: VDebugger! Discover how our novel framework debugs visual programs using execution feedback, boosting accuracy and interpretability by up to 3.2%!
Project:
Paper:
Code:
Looking for a debugging algorithm for visual programming? Take a look at 𝗩𝗗𝗲𝗯𝘂𝗴𝗴𝗲𝗿🔥🔥🔥
By tracking execution step by step, VDebugger boosts the accuracy by up to 𝟯.𝟮% on 6 visual reasoning tasks!
9
13
32
2
12
55
🤯So thrilled to have
@AnthropicAI
benchmark their latest, powerful Claude 3 models on our
#MathVista
for visual math reasoning!
It's encouraging to see the rapid progress in (multimodal) LLMs, especially in the math and science fields! 💥
🤗 Our
@huggingface
Data:
Today, we're announcing Claude 3, our next generation of AI models.
The three state-of-the-art models—Claude 3 Opus, Claude 3 Sonnet, and Claude 3 Haiku—set new industry benchmarks across reasoning, math, coding, multilingual understanding, and vision.
573
2K
10K
1
7
52
🔥Thrilled to see our
#LLaMA
-Adapter featured in
@HuggingFace
's "Spaces of the Week"! 🎉
Introducing LLaMA-Adapter V2, our cutting-edge multi-modal instruction model! Explore demo examples here: 💡
🚀Stay tuned for the technical report and model release!
0
10
51
🚀 Our
@Gradio
demo now supports diverse vision-language tasks:
1️⃣ Visual Question Answering (VQA)
2️⃣ Multi-level Dense Caption
3️⃣ Referring Expression Comprehension
4️⃣ Relationship Grounding
5️⃣ Grounding Captions
6️⃣ Object Detection
7️⃣ Human Keypoint Detection
8️⃣ Text Detection
0
11
48
It has been a wonderful day at Open House
@allen_ai
🍺🍖🌊. I met a lot of great people and got inspiring advice. Many thanks to the great efforts of the operations team for preparing all of it!
0
2
50
🎉 Exciting news! Our
#MathVista
is excelling with the latest advances in vision-language models (VLMs). Grok-1.5V by
@xai
achieves a 52.8% score, surpassing leading models such as GPT-4V, Claude 3 Opus, and Gemini Pro 1.5!
🔗 Visit our project page:
👀
1
4
46
Congratulations and thanks to
@MistralAI
for releasing the
#MoE
model to the community.
Our LLaMA2-Accessory now features Mixtral-8x7b with a chatbot demo, available on
@Gradio
!
Try the Chatbot:
http://106.14.127.192/
For more implementation details:
📖 Documentation:
0
10
43
📢 Attention
#NLPoc
community!
Submit and showcase your research at the 4th Southern California Natural Language Symposium (SoCal NLP) 📜
🗓️ Submission Deadline: Oct. 21, 2023, 11:59 PM PT
🔗 More info:
#SoCalNLP
#CallForPapers
1
13
45
Thanks for sharing our work! 🦙🍻
LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model
Compared to the original LLaMAAdapter, LLaMA-Adapter V2 can perform open-ended multi-modal instructions by merely introducing 14M parameters over LLaMA
abs:
github:
3
99
342
0
6
42
🚀 Excited to see Claude 3.5 Sonnet by
@AnthropicAI
achieve a new SOTA on
#MathVista
with 67.7%, a 19.8% improvement over Claude 3 Sonnet! 📈🎉
Learn more:
📝 Blog:
🔢 MathVista:
Introducing Claude 3.5 Sonnet—our most intelligent model yet.
This is the first release in our 3.5 model family.
Sonnet now outperforms competitor models on key evaluations, at twice the speed of Claude 3 Opus and one-fifth the cost.
Try it for free:
442
2K
7K
1
8
43
Gratitude to our esteemed speakers, insightful panelists, engaged attendees, and dedicated organizers (
@LiangZhenwen
,
@AlbertQJiang
,
@katie_m_collins
,
@KaiyuYang4
,
@wellecks
, and
@JLMcClelland
) for making the 3rd
#MATHAI
workshop at
#NeurIPS2023
an extraordinary success!!
📢 Can't wait to see you at the 3rd
#MathAI
Workshop in the LLM Era at
#NeurIPS2023
!
⏰ 8:55am - 5:00pm, Friday, Dec 15
📍 Room 217-219
🔗
📽️
Exciting Lineup:
⭐️ Six insightful talks by
@KristinLauter
,
@BaraMoa
,
@noahdgoodman
,
4
21
88
1
4
42
🚀We've just launched
#SciBench
, a sophisticated, college-level benchmark. It uniquely evaluates the capabilities of LLMs in tackling scientific problem-solving.
SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models
paper page:
Recent advances in large language models (LLMs) have demonstrated notable progress on many mathematical benchmarks. However, most of these
2
17
67
1
8
40
In 2021, we explored early research in geometry: our Inter-GPS, a neuro-symbolic solver, reached average human-level score for the first time.🎉
Now,
@GoogleDeepMind
's AlphaGeometry marks a historic breakthrough: Olympiad-level skill!🚀
🔎For more:
🔗
Introducing AlphaGeometry: an AI system that solves Olympiad geometry problems at a level approaching a human gold-medalist. 📐
It was trained solely on synthetic data and marks a breakthrough for AI in mathematical reasoning. 🧵
127
1K
4K
1
8
36
Happy to receive the NeurIPS 2022 Scholar Award! I really appreciate every support I get from the community, and I will devote myself to making contributions to the community!
@NeurIPSConf
🍻See you in New Orleans!
1
1
38
Still buzzing from the
#CopilotPCs
launch yesterday, and now
@Microsoft
drops the efficient Phi-3-Vision model! 🚀 Thrilled to see three of our past projects, featured in their benchmarks!
Encouraged to continue pushing the boundaries of AI research! 💡📊🔍
ScienceQA -
Phi-3-vision looking enticing.
128K context
4B parms
Performs exceptionally well on benchmarks
Will have to see if this one translates well to real-world use, but I am excited to check it out.
11
15
159
2
5
39
⭐️ Awesome!
@guyvdb
from UCLA is presenting the talk "AI Can Learn from Data. But Can It Learn to Reason?" offering insights from a logical and probabilistic perspective!
#MATHAI
#NeurIPS23
#Logic
#Reasoning
#AI
📢 Can't wait to see you at the 3rd
#MathAI
Workshop in the LLM Era at
#NeurIPS2023
!
⏰ 8:55am - 5:00pm, Friday, Dec 15
📍 Room 217-219
🔗
📽️
Exciting Lineup:
⭐️ Six insightful talks by
@KristinLauter
,
@BaraMoa
,
@noahdgoodman
,
4
21
88
0
3
37
🚨 Attention! I'm presenting the 🦎
#Chameleon
paper at Booth 320 from 10:45 to 12:45 at
#NeurIPS23
. You're welcome to stop by for a chat! ☕️😉🤖🧲💡
For more details, check out our project at .
Chameleon: Plug-and-Play Compositional Reasoning with Large Language Models
Chameleon with GPT-4 achieves an 86.54% accuracy on ScienceQA, significantly improving upon the best published few-shot model by 11.37%; using GPT-4 as the underlying LLM, Chameleon achieves a 17.8%
0
101
413
2
3
34
🚀
@google
is introducing new updates to aid in learning math and science, especially in visual contexts: .
💥 We're proud to spotlight our commitment to math and science over the past years, with projects like
#MathVista
,
#Chameleon
, and
#ScienceQA
.
1️⃣
0
10
33
It is remarkable that Gemini achieves a new SOTA of 53.0% on MathVista (), a challenging benchmark for math reasoning in visual contexts. We are honored that our proposed
#MathVista
is advancing the development of the newest and most capable AI models.
In image understanding, Gemini performs well across all the benchmarks we examined, with the Ultra model setting new state-of-the-art results in every benchmark.
4
9
192
0
3
34
🧲Please stop by our poster on deep learning for math reasoning at Poster Session 2
@aclmeeting
#ACL2023NLP
.
❤️Thanks to co-authors for their great contributions:
@liangqiu_1994
,
@wyu_nd
,
@wellecks
, &
@kaiwei_chang
.
abs:
github:
0
5
34
🚀OpenAI is releasing the latest function and tool-calling update for
#GPT4
!
Just two months back, we introduced
#Chameleon
🦎, an innovative compositional reasoning framework. It uses LLMs as a planner to generate diverse programs, integrating various tools including LLMs,
0
6
33
It was great to attend the
#NeurIPS2022
poster session and present our work
@UCLA
@ASU
@allen_ai
in person🎉. I’m excited that I met many great people and got countless insightful advice and comments. Thanks to everyone for your interest in our work!🍻
0
4
32
Thanks for sharing our latest work on multimodal procedural planning 🍻
Multimodal Procedural Planning via Dual Text-Image Prompting
abs:
github:
0
35
127
0
3
28
🎯It is time to submit your work on mathematical reasoning to the 2nd MATH-AI workshop!
As the workshop is non-archival, papers that are recently published or under review are allowed. ⏰The submission deadline is due on Sep 29⏰.
✅✅More information:
0
7
30
🎉🎉I am really happy that the 2nd MATH-AI workshop ended with such a big success. Very encouraged that so many people are interested in the domain and that the community is growing rapidly. Huge thanks to the speakers, panelists, and organizers! See you all at future events!!🍻
2
1
30
🎉 Exciting News! X-Accessory now welcomes a new addition - Mistral-MoE! 🌟
Discover it here:
🚀 Tap into the power of Mistral-MoE with our X-Accessory's robust framework, with the new features of inference and LoRA fine-tuning via model parallelism.
🌐
0
7
29
😜Looking forward to seeing you at the 1st Tool-Augmented Vision (TAVI) Workshop at
#CVPR2024
in Seattle.
🔍For more details, please visit the website:
0
4
29
🤔Naming things is hard!!
🦎
#Meta
's new work shares the same name as our NeurIPS 2023 paper from one year ago: Chameleon: Compositional Reasoning with LLMs.
Coincidence or great minds thinking alike? 😈 Dive into our work here:
Newly published work from FAIR, Chameleon: Mixed-Modal Early-Fusion Foundation Models.
This research presents a family of early-fusion token-based mixed-modal models capable of understanding & generating images & text in any arbitrary sequence.
Paper ➡️
27
204
951
3
2
29
We're dedicated to
#OpenSource
, confident that it will profoundly enrich the community.🌟
Thrilled to see our recent work, LLaMA-Adapter, and its subsequent developments positively impacting the community.🚀
Stay updated with continuous improvements: 📌
It was a great month for open source: So many LLMs came out that it's become quite overwhelming to keep track of it all.
So, in this month's Ahead of AI issue, I am sharing resources and research insights on the latest open-source LLMs & datasets!
13
128
546
0
7
26
🚨Call for Papers🚨 Submission to the
#NeurIPS2022
MATH-AI Workshop will be due on Sep 30, 11:59pm PT (2 days after ICLR😆). The page limit is 4 pages (not much workload🤩). Work both in progress and recently published is allowed. Act NOW and see you in
#NewOrleans
!🥳🥳🍻
0
9
26
One model to align multiple modalities. Looking forward to seeing the live demo.
OneLLM: One Framework to Align All Modalities with Language
paper page:
Multimodal large language models (MLLMs) have gained significant attention due to their strong multimodal understanding capability. However, existing works rely heavily on
5
69
253
0
4
25
An excellent blog on Controllable Neural Text Generation from
@lilianweng
! It's important to consider ways to reduce the hallucinations of LLMs and better reflect human intentions, especially given their current success and limitations.
👉
#ChatGPT
#LLM
0
3
26
Thrilled to join the live event, thanks to
@LightningAI
's kind invitation! 🌟 Peng and I will share the insights behind the LLaMA-Adapter series.
📅 event:
📚 abs-1:
📚 abs-2:
💻 code:
0
7
25
1
1
23
Excited to be at
#AAAI23
on-site! Can't wait to catch up with old friends and make new ones.
📢I'll give an oral presentation on
#ScienceQA
() at
@knowledgenlp
Workshop on Monday, Feb 13, 2:15-3:15 pm in Room 144B.
If you're around, let's grab a coffee!
0
1
24
📢📢Welcome to the 2nd
#MATH
-AI workshop tomorrow (Sunday, Dec 03) in Rooms 293-294 at
#NeurIPS2022
if you are interested in math reasoning and AI! There are 6 invited talks, 3 contributed talks, 1 poster session, and 1 panel discussion.
🪜Full program:
0
7
23
🧵1/6 Experience the magic of LLaMA-Adapter! Transforming real-world inputs like text, images, videos, audio, and 3D point clouds into engaging text. The reality you know, reimagined through AI.
🖼️📽️🔉🌐➕📝 ➡️➡️🦙➡️➡️ 📝
2
4
20
Excited to see the breakthrough achieved by
@Apple
's MM1 model, as evidenced by our
#MathVista
(), the comprehensive benchmark for math reasoning in visual contexts!
Few-shot mixed-resolution CoT: we can keep the strong few-shot capabilities learned from multimodal pre-training even after instruction-tuning: MM1-30B-Chat achieves 39.4 zero-shot on MathVista, but with eight-shot CoT mixed-resolution prompting we can achieve 44.4.
1
4
24
0
1
20
0
1
20
🧐Looking for a well-designed benchmark for mathematical reasoning? Lila 📜 is your next best option! 🥳🥳
Can a language model help you with your math homework? Not on its own, but maybe with the help of a Python interpreter!
In our EMNLP paper we present 📜 Līla and 🤖 Bhāskara, a math reasoning benchmark and model.
📄:
🔗:
1/🧵
5
38
214
0
3
18
Excited to organize the 2nd MATHAI workshop
@NeurIPSConf
with our great team❤️! The workshop will be in New Orleans🏙️ in person, on December 03, 2022. The submission is open now🧲!
#NeurIPS2022
🚨We are organizing the 2nd MATHAI workshop at NeurIPS!
Check it out if you're interested in AI for math, and machine reasoning in general🤯!
We have a great lineup of speakers & panelists!
See more in call for papers: 👇
3
30
150
0
2
18
🥳Trilled in New Orleans for
#NeurIPS
! This year, I will present one paper (ScienceQA) + 2 WS papers (PromptPG, Lila). And I am co-organizing the 2nd MATH-AI workshop!
☕️Excited to meet you! DM me if you want to grab a coffee and chat about MathAI, LLMs, and trustworthy NLP!!👇
1
1
17
Evaluating response quality with GPT-4, LLaMA-Adapter-V2 outshines ChatGPT. It triumphs over
#ChatGPT
in response quality, scoring 102%:100%! 🚀
2
4
14