🤖JSON-based Agents With Ollama & LangChain Learn to implement a Mixtral agent that interacts with a graph database Neo4j through a semantic layer This work by @tb_tomaz is great for a few reasons: - Shows how to build an agent with an OSS model - Shows how to build and use a… https://t.co/oZP24MER8Q https://t.co/hvxUVXDnay
— LangChain (@LangChainAI) Feb 29, 2024
LLM Human-Level Forecasting -A RAG system automatically searches for relevant information, generates forecasts, & aggregates predictions -On questions from competitive forecasting platforms published after LLM cut-offs, system can beat human forecasters https://t.co/fEDZuOHqdJ https://t.co/OBLLdkDSZ6
— John Nay (@johnjnay) Feb 29, 2024
Is this too far? 👀👀 This is by far the wildest Audio2Video I have ever seen called EMO: Emote Portrait Alive https://t.co/uCpjUCT6cT
— Salma (@Salmaaboukarr) Feb 28, 2024
First Sora AI videos, and now Emo turns your images into talking or even singing videos 🤯 Look at these insane examples (link at the end): 1. AI girl cross-actor performance https://t.co/4ja7JMgZbP
— Alvaro Cintas (@dr_cintas) Feb 28, 2024
AI is going to have a profound impact on the enterprise. Software and IT teams in companies have been responsible for enabling tech for people to do their work. Now with AI, they’ll be providing the work itself. This is a fundamental shift in the power structure in companies.
— Aaron Levie (@levie) Feb 28, 2024
A new @kaggle competition called “AI Assistants for Data Tasks with Gemma” was launched last week! So, I built a quick Retrieval-Augmented Generation (RAG) pipeline to find/recommend the right Kaggle competition to review and learn from using: ✨ @Google's open source LLM Gemma… https://t.co/Ok7iBkoFlY https://t.co/stT4pTLkyI
— Leonie (@helloiamleonie) Feb 28, 2024
The difference between turbojet vs turbofan [🎞️ techsight] https://t.co/fi7FQK4HsQ
— Massimo (@Rainmaker1973) Feb 27, 2024
Is DNA all you need? In new work, we report Evo, a genomic foundation model that learns across the fundamental languages of biology: DNA, RNA, and proteins. Evo is capable of both prediction tasks and generative design, from molecular to whole genome scale. https://t.co/BPo9ggHhmp
— Patrick Hsu (@pdhsu) Feb 27, 2024
Another great use of AI: https://t.co/Bu6YsbK8aq
— Vinod Khosla (@vkhosla) Feb 27, 2024
Diffusion Transformer (DiT) is the foundation behind AI models like Sora. Just came across OpenDiT, an easy-to-use, fast, and memory-efficient system for DiT training and inference. OpenDiT is an open-source, high-performance implementation of DiT to enhance the efficiency of… https://t.co/4XP1UbQwrL https://t.co/6JgaUhyXYs
— elvis (@omarsar0) Feb 27, 2024
This is a mind-blowing announcement. Mistral AI, the French company that has been fighting tooth and nail to water down the #AIAct's foundation model rules, is partnering up with Microsoft. So much for 'give us a fighting chance against Big Tech'. A 🧵1/8 https://t.co/WJtKNkq1K8
— Luca Bertuzzi (@BertuzLuca) Feb 26, 2024
Demis Hassabis admits that Google has some secret sauce in how Gemini is able to process 1-10m token context windows. The extreme context length in Gemini 1.5 Pro "can't" be achieved "without some new Innovations". This is an astonishing development that seems to hint at… https://t.co/rzy0VbXEo2 https://t.co/8nQiGdyaNR
— Carlos E. Perez (@IntuitMachine) Feb 25, 2024
✳️Kubernetes Interview: What to Know✳️ https://t.co/Ndl6uhfXh0 https://t.co/UVsJPQHTUD https://t.co/Y7dDUHVgo7
— Tech Fusionist (@techyoutbe) Feb 26, 2024
Bill Ackman on dealing with disaster: “Make a little progress every day” After losing $4 billion in Valeant Pharmaceuticals, other Wall Street investors started betting against his firm, Pershing Square, to further compound its catastrophic 30%+ loss and force a liquidation. To… https://t.co/l7ZV6IENy2 https://t.co/jOiCnxWR0k
— Startup Archive (@StartupArchive_) Feb 26, 2024
@jenny____ai https://t.co/xL84vanQsw
— Yohei (@yoheinakajima) Feb 26, 2024
The first Skynet drones just dropped https://t.co/iuNWmcuYgM
— Jenny AI (@jenny____ai) Feb 26, 2024
The Q4 letter came out from Goehring & Rozencwajg. As always, an interesting read for anyone interested in natural resource markets. The essay on EVs was particularly interesting. Full letter: https://t.co/ANJB7lrb1c https://t.co/DL7VOHt173
— Dalius - Special Situation Investments (@InvestSpecial) Feb 26, 2024
Rare Peter Thiel interview on Thiel Fellowship in WSJ "We don’t believe education systems are working but that doesn’t mean we have a better way of education, so we’ve leaned into identifying talent and letting them figure it out." "“If you scale the program, “you will have a… https://t.co/MXg3i7bLH8 https://t.co/zDJyFKNTY3
— Sar Haribhakti (@sarthakgh) Feb 26, 2024
OpenAI going after synthetic data market? “Data labeling and annotation services, namely, data processing and systemization services using automated labeling and annotating of images, audio, video, text and other forms of electronic data …” https://t.co/Ht9X98YO3b https://t.co/SDeGkfO8gi
— Delip Rao e/σ (@deliprao) Feb 25, 2024
Just finished reading this book and would highly recommend 👇🏻 https://t.co/I6KGrNPdRP
— Leandro (@Invesquotes) Feb 25, 2024
Let’s walk through RAG pain points and solutions! 🧑🏫🎬 We’re excited to feature @wenqi_glantz for a video walkthrough video of her popular “12 RAG Pain Points and Solutions” blog post, which is the most comprehensive cheatsheet we’ve seen of pain points that occur at every stage… https://t.co/zwiOYLvmIV https://t.co/g0gZ7gtnuL
— LlamaIndex 🦙 (@llama_index) Feb 25, 2024
🪩Reflection Agents Agents are unreliable. One of the best ways to improve their accuracy is to run some step to check or critique their previous response. We've added three examples of this (based on recent papers): Reflexion Paper: https://t.co/UzClW9hNGI Language Agent Tree… https://t.co/bWwL7a4Ebz https://t.co/YawJZIxoND
— LangChain (@LangChainAI) Feb 24, 2024
Incredible tribute of Charlie Munger by Warren. A true friend. https://t.co/GOxBLFFTot
— tyler hogge (@thogge) Feb 24, 2024
The mobile S-curve ends, and the AI S-curve begins There’s never been a bigger contrast between mobile and AI — it’s the end of one technology curve, and the start of the other. It’s been 15 years since the App Store was launched; while the generative AI revolution started… https://t.co/9f4D2N6rhy https://t.co/5UlHYcmhJc
— andrew chen (@andrewchen) Feb 22, 2024
Polling Vs Webhooks - Polling Polling involves repeatedly checking the external service or endpoint at fixed intervals to retrieve updated information. It’s like constantly asking, “Do you have something new for me?” even where there might not be any update. This approach is… https://t.co/b6S0po3n4y https://t.co/CmCVa3hgk6
— Alex Xu (@alexxubyte) Feb 23, 2024
This is a fun debate to have. I could argue it both ways. @jaygoldberg https://t.co/UNpmuwrrgL
— Ben Bajarin (@BenBajarin) Feb 23, 2024
This is nearly always optimal. Same for design. https://t.co/I9MY41vhzb
— Steven Sinofsky (@stevesi) Feb 24, 2024
Of course. Nietzsche called it "leveling". The 20th century called it "communism". The relentless drive to the equality of the lowest common denominator. https://t.co/WFWE1yk3fI
— Marc Andreessen 🇺🇸 (@pmarca) Feb 23, 2024
A Visual Guide to CI/CD 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 (𝗖𝗜) is a foundational practice in DevOps where developers frequently merge code changes into the main branch, often multiple times a day. This process is complemented by automated testing to ensure new… https://t.co/nUnBe9eSxI https://t.co/751oOqEate
— Sahn Lam (@sahnlam) Feb 23, 2024
Tokenization is all the rage today (cc @karpathy's minbpe). In Tokenization and the Noiseless Channel we hypothesized that the higher the unigram Rényi entropy the better the tokenization and thus model performance. (entropy is a much better metric than compression length) https://t.co/02QIHKOxPU
— Vilém Zouhar (@zouharvi) Feb 23, 2024
Steve Martin "Be so good they can't ignore you" advice https://t.co/85x8mCu2Z5 https://t.co/W08Vk3DPrW
— Startup Archive (@StartupArchive_) Feb 24, 2024
Companies often begin in opposition to something — a bad incumbent competitor, bad culture ideas, or just a bad status quo. They then work feverishly to become that which they opposed.
— Marc Andreessen 🇺🇸 (@pmarca) Feb 23, 2024
Did someone say they were worried about @Groq's cost/ability to scale? 👀 #GroqSpeed https://t.co/IYx6CGOW3o
— Jonathan Ross (@JonathanRoss321) Feb 23, 2024
Introduction to Matryoshka Embedding Models in Open Source!🪆 The embedding models from @OpenAI use a concept called Matryoshka that allows you to dynamically reduce the dimensions of your embedding to save storage and decrease latency without sacrificing much performance. 💰🏎️… https://t.co/vzVXsbUPMX https://t.co/aTKBb4XJgc
— Philipp Schmid (@_philschmid) Feb 23, 2024
📨 I put the Satoshi emails into Gemini 1.5 Pro, and asked if it could help us infer anything about their location, identity, personality, or gender. 👇Results below! https://t.co/3wezY494Yc https://t.co/D7jON24Dcz
— 👩💻 Paige Bailey (@DynamicWebPaige) Feb 24, 2024
pod pod pod!!! https://t.co/kt0vT3JzXK
— Chamath Palihapitiya (@chamath) Feb 24, 2024
Matryoshka Embeddings are here! 🔥 The Sentence Transformers library allows training and running embedding models with embedding sizes that can be shrunk while keeping high quality! Learn about them in https://t.co/M5defsp8o4 https://t.co/nDuteMA39N
— Omar Sanseviero (@osanseviero) Feb 23, 2024
I just got access to Gemini 1.5 Pro and recorded a video for you. I'm in shock. I asked Gemini to ... 🤓 Read every single Sherlock Holmes book (1,221 pages or 919,243 tokens) 🔍 Find two relevant quotes for my presentation ✅ Output exact quoted relevant text 🧠 Summarize… https://t.co/leYS3K6RZb https://t.co/ByBP87HDjy
— Ryan Carson (@ryancarson) Feb 21, 2024
I fed Google Gemini 1.5 Pro the last 8 quarters of Amazon shareholder reports and call transcripts. Wow. "What was an Amazon focus for 2022 that is weirdly absent from the 2023 shareholder calls and reports?" https://t.co/iboXm708t5
— Allie K. Miller (@alliekmiller) Feb 22, 2024
Breaking New Ground: Lag-Llama Redefines Time Series Forecasting The paper introduces Lag-Llama, a new LLM for probabilistic time series forecasting. While LLMs have shown success in natural language processing and computer vision, their development has lagged for time series… https://t.co/wkz3v8qt63 https://t.co/tMBwz5Lj6H
— Carlos E. Perez (@IntuitMachine) Feb 21, 2024
oh hello, @karpathy has dropped a two hour video on tokenization, the “real root of suffering” https://t.co/ln9C1HXjrx https://t.co/umkwiA8hBz
— morgan — (@morqon) Feb 20, 2024
Using @GroqInc to instantly refactor and document code. The implications for software engineering are wild. Gone are the days of waiting on an LLM for suggestions or code changes. Now, it's an instant feedback loop. Demo link in the comments: https://t.co/EB3pVwEuaN https://t.co/zPdMPJ0Uc1
— Matt Shumer (@mattshumer_) Feb 19, 2024
Highly recommend watching this talk by @karpathy to build an understanding of how LLMs work https://t.co/clo6iaA12X https://t.co/85xdlVekq9
— Lenny Rachitsky (@lennysan) Feb 19, 2024
A key insight for advanced RAG is that you need to think about how to optimize all stages of the retrieval pipeline, from pre-retrieval (how do you ingest/represent the data) to retrieval (hybrid search) to post-retrieval (reranking). This is a fantastic blog post and diagram by… https://t.co/qNWzEuATBC https://t.co/NERv4FBkJe https://t.co/qK9Mh2t2Kn
— Jerry Liu (@jerryjliu0) Feb 20, 2024
its just been a month since this post but the GPU Rich War has a new leader - @GroqInc has come out from nowhere* to completely tilt the economics of inference. Only provider in green square on latency vs throughput vs price, because they are not just timesharing H100s like… https://t.co/tBAdfHopNF https://t.co/h46jgKhpgk https://t.co/WNJq2hNqcT
— swyx (@swyx) Feb 20, 2024
Demos from AI chip maker Groq go viral after the company's inference engine shows remarkable speeds running LLMs, including with real-time conversations (@zeffmax / Gizmodo) https://t.co/C9cj7drcQh 📫 Subscribe: https://t.co/OyWeKSRpIM https://t.co/r4hHahtvLy
— Techmeme (@Techmeme) Feb 20, 2024
✨ Very nice paper indeed to know the current state of RAG - "Retrieval-Augmented Generation for Large Language Models: A Survey" 📌 The paper classifies Naive RAG, Advanced RAG and Modular RAG. 📌 Breaks down the Advanced RAG techniques into Pre-retrieval, Retrieval, and… https://t.co/j0RfBlUJxV https://t.co/WWSngtdb5l https://t.co/rxD9jAUkOi
— Rohan Paul (@rohanpaul_ai) Feb 20, 2024
New @karpathy video just dropped! 😍🔥 After watching, if you want to learn more about how different models (e.g., GPT4, Llama, T5, BERT) tokenize text, check out "The Tokenizer Playground": a web-app I built a few months ago with 🤗 Transformers.js! 🔗 https://t.co/1AhHmbXpsu https://t.co/X4nqi3Nd2D https://t.co/vTD2niiD0V
— Xenova (@xenovacom) Feb 20, 2024
We've got a looming crisis here! Advanced RAG is going to grow in ridiculous complexity. We don't have the proper abstractions to deal with this kind of complexity. https://t.co/SlHg6xBdjD
— Carlos E. Perez (@IntuitMachine) Feb 20, 2024
Just tried LlamaParse on LlamaCloud 🦙 Gave it a 249-pages snippet of the Singapore government's FY24 budget - this contains hundreds of tables with hundreds of line items. Congratulations @llama_index @jerryjliu0 and the team on the launch! https://t.co/m0UrZMuobZ
— gabriel (lu.ma/sg-ai) (@gabchuayz) Feb 20, 2024
Man I love Karpathy! https://t.co/FxpVLwJULK
— rohit (@krishnanrohit) Feb 20, 2024
Groq is a Radically Different kind of AI architecture Among the new crop of AI chip startups, Groq stands out with a radically different approach centered around its compiler technology for optimizing a minimalist yet high-performance architecture. Groq's secret sauce is this… https://t.co/10XjAcUc37 https://t.co/Z70sihHNbx
— Carlos E. Perez (@IntuitMachine) Feb 20, 2024
The problem with "Cosine similarity" for vector databases - very nicely explained in this piece from @JinaAI_ "Many practical vector databases and neural search solutions depend on fast cosine similarity matching between document and query embeddings. While appealing for its… https://t.co/PDCxj0WCCg https://t.co/1TBSrYK39C
— Rohan Paul (@rohanpaul_ai) Feb 20, 2024
Some of my favorite @naval ideas: 1. If you don't know yet what you should work on, the most important thing is to figure it out. 2. Books make for great friends, because the best thinkers of the last few thousand years tell you their nuggets of wisdom. 3. You will get rich… https://t.co/UnB61tzw9e
— David Senra (@FoundersPodcast) Feb 20, 2024
Gemini 1.5 might be viewed as one of those watershed moments in history. Google was one of 15 search engines back in the early 2000s and quietly rose to #1 beating everyone else. They know better than anyone how to get to the leadership position https://t.co/I9LNV2gX3h
— Anu Hariharan (@anuhariharan) Feb 19, 2024
Until now, HELM has evaluated LMs with on short responses, where evaluation is simple. We now introduce HELM Instruct, which evaluates open-ended instruction following. We evaluate 4 models on 7 scenarios using 4 evaluators against 5 criteria: https://t.co/5dyRHDjlrF
— Percy Liang (@percyliang) Feb 19, 2024
Q* from OpenAI and tree-of-thought reasoning triggered a lot of enthusiasm on augmenting LLMs' reasoning/planning capabilities with search. But is search really the panacea for LLMs? Answer from our new study @osunlp: Not quite yet. TLDR: For advanced planning methods like tree… https://t.co/jq7rqv1ubB https://t.co/UeYaxONuOw https://t.co/nORMFc2Xsx
— Yu Su (@ysu_nlp) Feb 20, 2024
Vector Databases clearly explained!🔥 Vector Databases are really key when it comes to working with LLMs which helps in faster retrieval and process of the data. If you are not sure what exactly it is, here's a detailed introduction. Let's start with Vector Embeddings:🚀… https://t.co/0KZ7OjHsIW https://t.co/Y6ikJZUusi
— Sumanth 🚀 (@Sumanth_077) Feb 19, 2024
Comparison between RAG and Fine-tuning From paper - "Retrieval-Augmented Generation for Large Language Models: A Survey" https://t.co/2l8Aeq7YD6
— Rohan Paul (@rohanpaul_ai) Feb 19, 2024
How to improve API performance? Here are 5 tips: 1. 𝗣𝗮𝗴𝗶𝗻𝗮𝘁𝗶𝗼𝗻 Pagination is a common optimization technique when result sets are large. By streaming results back to the client in pages, service responsiveness can be improved. 2. 𝗔𝘀𝘆𝗻𝗰𝗵𝗿𝗼𝗻𝗼𝘂𝘀… https://t.co/EOuOwAys7c https://t.co/jCu49fGQDN
— Sahn Lam (@sahnlam) Feb 19, 2024
My latest guided project on focused on the large text summarization process using the popular #GenerativeAI framework @LangChainAI is available on the @coursera platform now. Spend less than 2 hours to learn a new #ai concept https://t.co/Km1sPLz9Rr
— datasciencewithsam👨💻🔰 (@soudey123) Feb 17, 2024
Finest summary I’ve read on the VisionPro, from an expert. In just two pages, you get the complete overview, in 1 minute. Then, the next step is to go experience it firsthand. 🤩 https://t.co/i37iPetFYQ
— Emm (@emmanuel_2m) Feb 17, 2024
I built a very simple demo for an alien creature merging/breeding game with infinitely varied and never-ending content using Scenario and a strongly typed LLM function framework that we’re tinkering with at Banana. Key steps in thread below, as well as the link to try it. 🧵👇 https://t.co/CZKEzS45QI
— Nik 🪐 (@entreprenik) Feb 17, 2024
Fiction books to AI Movie generation: A rough blueprint of how fiction books can be made into movies using AI in the near future using tools available by @LangChainAI @UnstructuredIO @elevenlabsio @replicate @midjourney Maybe I could try building a quick MVP, let’s see. https://t.co/Aj90ScIir9
— Misbah Syed (@MisbahSy) Feb 18, 2024
Gemini 1.5 Pro's 1,000,000+ token context length is incredible. I got early access and spent my Saturday night running tests. Here are 6 impressive capabilities I found: https://t.co/Rh8qGCUevE
— Rowan Cheung (@rowancheung) Feb 18, 2024
Don't use Upwork and Fiverr. Because they're saturated. Here are top 10 sites to get a remote job that pays in USD: https://t.co/6LYubpJMHp
— Ihtesham Haider (@ihteshamit) Feb 17, 2024
Interesting leadership principle by Zuck shared on Morning Brew's podcast today: "probably one of my most controversial leadership or management things is I don't actually believe in delegating that much. I kinda think like the way a founder should work is you should basically… https://t.co/Cqjb7sipyA
— Mostly Borrowed Ideas (@borrowed_ideas) Feb 16, 2024
Back in 1984, as Apple was already public and they were pushing the Macintosh, Steve Jobs highlights the problem with hiring outside management. https://t.co/ywOFQ0SKl5
— Historic Vids (@historyinmemes) Feb 17, 2024
RAG From Scratch: Query Translation (RAG-Fusion) Our RAG From Scratch video series walks through impt RAG concepts in short / focused videos w/ code. This is the second in our videos on Query Translation, focused on RAG-Fusion by @Raudaschl. 🔧 Problem: A single user… https://t.co/LjSjuSgpKb https://t.co/FEZH5hUktS
— LangChain (@LangChainAI) Feb 16, 2024
Just went over Gemini 1.5's technical report (58 pages). One part totally blew my mind. The model achieves near-perfect “needle” recall (>99.7%). It’s a HUGE deal. LLMs often drop the ball on remembering stuff. The more they have to remember, the more they forget. Imagine… https://t.co/KDXuAWAkiT https://t.co/xwtceIic3v
— Lior⚡ (@AlphaSignalAI) Feb 16, 2024
Pretty much, just describe to edit. In the sora technical paper they note that they apply SDEdit by @chenlin_meng et al fine - you can take that further through scene decomposition, segmentation, regional prompting & more to autocorrect & balance image, video, audio, etc etc https://t.co/C2k4vf2bkm https://t.co/hzpQcZpeVg
— Emad (@EMostaque) Feb 17, 2024
✅ Corrective RAG with LangGraph As @RLanceMartin once said “don’t build agents, build graphs” We’ve just implemented 4 new notebooks outlining different RAG and CRAG techniques in @LangChainAI PY & JS! These show off different RAG flows, using OSS and hosted LLMs. See the… https://t.co/ezPFLQKuQW https://t.co/tjAmNJWV3y
— Brace (@BraceSproul) Feb 16, 2024
The toolbox of AI alchemy (or metallurgy) keeps growing! Quantization LoRa wrapping Model merging Mixture of Experts (MoE) Extending layers What other methods did I miss that involve just reconfiguring the architecture and weights of a neural network?
— Carlos E. Perez (@IntuitMachine) Feb 16, 2024
OpenAI unveils its first text-to-video model, Sora, which can create up to a minute of 1080p video, as a research product for some creators and security experts (@stevenlevy / Wired) https://t.co/cRv0OwZGlv 📫 Subscribe: https://t.co/OyWeKSRpIM https://t.co/sanzXhpw3n
— Techmeme (@Techmeme) Feb 15, 2024
I'm gone for 12 hours and OpenAI just tossed the whole world upside down again 🤯 Even though I'm keeping up with all the Generative AI progress, Sora is such a big jump in video generation capabilities that it feels unreal. Time for a mega thread 🔽 https://t.co/OdhrB8LPPE
— Dreaming Tulpa 🥓👑 (@dreamingtulpa) Feb 16, 2024
I’ve been using LangSmith for over 6 months and it’s really great. You don’t have to instrument your code - all you do is set an env variable and the logging begins Logging to something like langsmith is key if you want to eventually fine-tune and build an eval harness. Also… https://t.co/iieBpGPvXr https://t.co/spiS9sI1Zd
— Hamel Husain (@HamelHusain) Feb 15, 2024
Generative 3d environments just became a thing with the announcement of OpenAI's new video model, Sora. Michael Rublof from https://t.co/vQKnaqIb5G took one of those videos, and turned it into a NeRF using Colmap and Nerfstudio. While people are laughing at the topology of… https://t.co/2y7lsbrdW2 https://t.co/8i96wmPjSO
— Martin Nebelong (@MartinNebelong) Feb 15, 2024
This is incredible. A collection of people stumbling upon Founders! https://t.co/YLFintlfIt
— David Senra (@FoundersPodcast) Feb 16, 2024
this is very exciting for anyone who is working on large scale software projects. Automated valuable unit test creation is an excellent near term contribution vector for llms. I imagine that this can become very powerful driver of higher quality software https://t.co/d4SXbUBXP5
— tobi lutke (@tobi) Feb 15, 2024
Saw @RLanceMartin do a demo of LangGraph on the @ollama meet-up. The idea of isolated nodes in a graph is something I have been playing around with for a while as an alternative to agent based reasoning. This very well executed by the @LangChainAI team imo. Keep shipping! https://t.co/EEE9kPSjfp
— homanp (@pelaseyed) Feb 13, 2024
📽️Editing LLM knowledge is possible, e.g., Rank-One Model Editing (ROME). 📔Paper: https://t.co/ZtKlyFnR2B 🎯Sample code: https://t.co/oWYnKP7mG1 💣The technology behind looks interesting and useful, which is supposed to work with SFT and RAG to reduce the hallucination!
— Haihao Shen (@HaihaoShen) Feb 15, 2024
Gemini 1.5 Pro - A highly capable multimodal model with a 10M token context length Today we are releasing the first demonstrations of the capabilities of the Gemini 1.5 series, with the Gemini 1.5 Pro model. One of the key differentiators of this model is its incredibly long… https://t.co/gpz7pMFFDX https://t.co/2KLro4VwLT
— Jeff Dean (@🏡) (@JeffDean) Feb 15, 2024
Current works are restricted to short sequences of texts and images, limiting their ability to model the world. Presenting Large World Model (LWM): capable of processing long text, images, videos of over 1M tokens (and *no* lost in the middle!) Project: https://t.co/IyHShZbXST https://t.co/5cd17UIcCk
— Pieter Abbeel (@pabbeel) Feb 14, 2024
Here's how the new "memory" feature of ChatGPT works: it uses a new tool called "bio" described in the system prompt like this: https://t.co/m5eWPBTi6N
— Simon Willison (@simonw) Feb 14, 2024
"One of the things that made Apple great is that, in the early days, it was built from the heart - not by somebody who came in and said 'I know how to build a company. Here's what you do: TATATATATA.' It wasn't built that way. It was built from the heart." — Steve Jobs, 1985
— Steve Jobs Stories (@SJobs_Stories) Feb 14, 2024
Jony Ive explains why Steve Jobs believed the iPhone had to come before the iPad: "He had always been leaning toward making a tablet. But shortly after he returned from surgery, during one of their regular brainstorming walks around the Apple campus, Steve told Jony Ive… https://t.co/KOVcv8XjQ6 https://t.co/SfNBfZXb0p
— David Senra (@FoundersPodcast) Feb 14, 2024
Naval Ravikant: "You should have a definite view of the future. You should actually go out and build the thing you want to see exist, rather than sit back and wait for somebody else to do it. History doesn't work that way. Things don't just automatically show up. There are… https://t.co/vs5K8O4aPz https://t.co/GaWa3eVC80
— Brian Ji (@BrianJJi) Feb 14, 2024
Build Agentic Workflows from Scratch 🤖🔁 It’s tempting for an AI engineer to use an off-the-shelf agent, but building your own agents shouldn’t be scary (and limited to AI research papers)! In our brand-new video tutorial we show you how to build an agent in two intuitive… https://t.co/2wZY1QNIC6 https://t.co/Oq6p80gcop
— LlamaIndex 🦙 (@llama_index) Feb 14, 2024
New LlamaIndex Webinar 🌟: Build No-Code RAG @FlowiseAI is one of the leading no-code tools for building LLM-powered workflows. We're excited to host @henryhengzj, co-founder of @FlowiseAI this Friday in our LlamaIndex Webinar series. With our recent LlamaIndex.TS + Flowise… https://t.co/IR0XvnE7qW https://t.co/RfVOBuVzTz
— LlamaIndex 🦙 (@llama_index) Feb 13, 2024
David Sacks: If you’re going to have a breakout startup, you need to innovate on distribution “If you're going to have a breakout startup, you've really got to think about how you're going to innovate on distribution, not just product. The biggest mistake I see these days is… https://t.co/sIugiER8oF https://t.co/Nq8w7FAKUu
— Startup Archive (@StartupArchive_) Feb 13, 2024
A recent survey on Retrieval-Augmented Generation (RAG) mentions an evolving paradigm: Modular RAG. Modular RAG is comprised of various functional modules. Thus, modular RAG is not standalone. Instead, different RAG patterns are composed of different modules. For example,… https://t.co/7GHR1TcmQ8 https://t.co/zI6MJsa0ya
— Leonie (@helloiamleonie) Feb 13, 2024
Yesterday's ETF inflows (source @FarsideUK) It was "only" $493 million, need to pump those numbers up, those are rookie numbers. $374 million by Blackrock. Biggest surprise might be the 2nd negative day by Invesco Galaxy Bitcoin ETF $BTCO. $GBTC with $95 million outflows… https://t.co/qxZVijDmhb https://t.co/IqJFjEYWjm
— WhalePanda (@WhalePanda) Feb 13, 2024
7 Key Metrics to Evaluate RAG Pipelines with LLMs RAGs leverage external data to enrich the context of LLMs, thereby enhancing their ability to generate more accurate and relevant responses. As the adoption of RAGs grows, so does the complexity of evaluating their performance… https://t.co/ea3iTN6App https://t.co/nRYBYbe0ba
— Shubham Saboo (@Saboo_Shubham_) Feb 12, 2024
Silicon Valley Has a Harvard Problem by Alexandar Karp of Palentir https://t.co/Q4y7jqkypZ https://t.co/5jKVloKOra
— Steven Sinofsky (@stevesi) Feb 13, 2024
Google has introduced VideoPoet, and it's MIND-BLOWING. It's literally ChatGPT for: - Text to video - Image to video - Video editing Here are 10 jaw-dropping examples:↓ https://t.co/mVXPbCLGk8
— Rishabh (@Rixhabh__) Feb 12, 2024
Jensen Huang just called for all countries to develop 'Sovereign AI' at the World Government Summit in Dubai. In other words, data is the new gold, and every country needs to own the production of its own artificial intelligence. "You cannot allow that to be done by other… https://t.co/LnK9IWWmgL https://t.co/f5gK5uqp5V
— Rowan Cheung (@rowancheung) Feb 12, 2024
(1/7) The G in RAG Trips Up GPT-4 .... Or Does it? The retrieval stage of RAG gets an incredible amount of attention. The generation stage ... not so much 🤔 We set out to test how models handle the Generation stage of RAG .. and @AnthropicAI beat @OpenAI #GPT4 ‼️ Most Evals… https://t.co/LmgfxIHlL2 https://t.co/NiYBrujzLh
— Aparna Dhinakaran (@aparnadhinak) Feb 12, 2024
When Barry Sternlicht was 32 y/o he started Starwood with Bob Faith and an assistant. 18 months later, after acquiring 8,000 MF units, he sold to Sam Zell’s Equity Residential REIT in return for 20% ownership. By the time he was 37 they owned Starwood Hotels and Lodgings Trust,… https://t.co/aEo0nO0Mus https://t.co/s8Dlhzo9f5
— Chris Powers (@fortworthchris) Feb 6, 2024
9/ LLMs for Table Processing - provides an overview of LLMs for table processing, including methods, benchmarks, prompting techniques, and much more. https://t.co/6t3kBVTISA
— DAIR.AI (@dair_ai) Feb 11, 2024
Hear, hear, AMD MI300Xs have started to emerge much sooner than expected. Here is a 2-part benchmarks report on performing BLOOM-176B inference using @MSFTDeepSpeed optimized for AMD MI300X. 1. https://t.co/JIxZj2U1NU 2. https://t.co/XdLgoB0ZOK This was published in response… https://t.co/nfpkpk0AAy
— Stas Bekman (@StasBekman) Feb 12, 2024
If anyone needs a relaxing, peaceful prompt, here's one: cinematic shot, DJI Mavic 3, organic architecture, photography, northern ireland, cold weather, overcast, light blue and green color palette, cinematic, raining Upscaled in Magnific https://t.co/YoziUzg5i5 https://t.co/dBCBaU9uLt
— Everett World (@WorldEverett) Feb 11, 2024
⬆️🏬 Upstash Vector Store One of the most important parts of any RAG pipeline is the vector store. To help with that, @LangChainAI 🦜🔗JS/TS now offers support for @upstash’s new vector store offering! Try it out today via the `@langchain/community` package. https://t.co/iPd4gi1Kct
— Brace (@BraceSproul) Feb 10, 2024
For years, I was hyperparameter tuning XGBoost models wrong. In 3 minutes, I'll share one secret that took me 3 years to figure out. When I did, it cut my training time 10X. Let's dive in. 1. XGBoost: XGBoost (eXtreme Gradient Boosting) is a popular machine learning algorithm,… https://t.co/uyhopiuuTN https://t.co/JMxKRfkZnB
— 🔥 Matt Dancho (Business Science) 🔥 (@mdancho84) Feb 11, 2024
Microsoft just dropped: An Interactive Agent Foundation Model "The development of artificial intelligence systems is transitioning from creating static, task-specific models to dynamic, agent-based systems capable of performing well in a wide range of applications. We propose an… https://t.co/wKmT8cZtN0 https://t.co/zMg5zrAH55
— Burny — Omni/Acc (@burny_tech) Feb 10, 2024
Three-levels of Text-to-SQL 🪜 Building sophisticated text-to-SQL is way more than a simple prompt, and actually you can combine RAG concepts into the text-to-SQL process to enabling querying over a large collection of tables in a robust manner. Here’s a brand-new tutorial that… https://t.co/ijwbDbV49Y https://t.co/heDd5Dr2g7 https://t.co/OJWTaLxQY7
— Jerry Liu (@jerryjliu0) Feb 11, 2024
Stripe founder John Collison on the importance of transparency “All of the things people talk about like hiring really great people or giving them a huge amount of leverage—transparency, for us, plays into that… We think that if you are aligned at a high level about what Stripe… https://t.co/yKwZiucpyJ https://t.co/ZSJSaE0zHv
— Startup Archive (@StartupArchive_) Feb 10, 2024
Finetuning LLM for domain-adaption https://t.co/QcTVAQvmAV
— /MachineLearning (@slashML) Feb 10, 2024
Some Qs to ask when building an autonomous agent, around topic of… Task planning and mgmt Task execution and routing Tool usage and learning Memory and knowledge Self improvement UI/UX (input/output) Auth and tool auth Personalization & generalization Security Let’s dig into… https://t.co/sZyeFS4zZS
— Yohei (@yoheinakajima) Feb 9, 2024
Based on playing with both ChatGPT 4 and Gemini Advanced side by side recently, I must say that both are quite impressive. One difference I noticed though is that GPT 4 tends to be more technical. Gemini Advanced often produces more natural sounding text though (no "delving"… https://t.co/KDWVErGJT4 https://t.co/YDCsKDg6Wd
— Sebastian Raschka (@rasbt) Feb 9, 2024
Is Online AI feedback (OAIF) the next iteration of DPO/RLHF? OAIF utilizes an LLM to provide online feedback, demonstrating superior performance over offline DPO and RLHF methods through human evaluation. 🤔 Online, in RLHF/RLAIF refers to the training data being acquired… https://t.co/a6uCgSL2kb https://t.co/L0Smp6sCbE
— Philipp Schmid (@_philschmid) Feb 9, 2024
Sources: OpenAI's annualized revenue hit $2B in December, up from $1.3B in October 2023, and the company believes it can more than double the figure in 2025 (Financial Times) https://t.co/0iIpoPjjCc 📫 Subscribe: https://t.co/OyWeKSRpIM https://t.co/akTJa2s0tP
— Techmeme (@Techmeme) Feb 9, 2024
Create an AI influencer in 5 minutes. Meet Jade. 👇 https://t.co/kntFwQnFqi
— Alie Jules (@aliejulesai) Feb 8, 2024
Come learn about how to reason about tabular data with LLMs. We've brought in some of the leading researchers on this topic (@zlwang_cs and @ltyleoii22). There's still time to signup, happening tomorrow at 9am PT: https://t.co/p6LkVHwIYU https://t.co/naabgGSEZF
— Jerry Liu (@jerryjliu0) Feb 9, 2024
An excellent, in-depth look at how to use LlamaIndex's new Query Pipelines feature to build complex retrieval strategies like sentence window retrieval, hybrid fusion retrieval, long context reordering and query rewriting in a more declarative way: https://t.co/Funqm7Jw1u https://t.co/CSd1jxu7zD
— LlamaIndex 🦙 (@llama_index) Feb 7, 2024
Corrective/Self-Reflective RAG in LangGraph Self-reflection w/ RAG is a cool idea from a few recent papers - Self-RAG (@AkariAsai et al), CRAG, etc. I tried laying out the flows from each paper as graphs and works pretty well. Short vid w/ code links: https://t.co/TcxaAoKf8D https://t.co/wHTzcjCuyF https://t.co/GVfsVfry5p
— Lance Martin (@RLanceMartin) Feb 7, 2024
The U.S. Now Has: 1. Record $17.5 trillion in household debt 2. Record $12.3 trillion in mortgages 3. Record $1.6 trillion in auto loans 4. Near record $1.6 trillion in student loans 5. Record $1.1 trillion in credit card debt Total mortgage debt is more than double the… https://t.co/9KoWj9jYCX
— The Kobeissi Letter (@KobeissiLetter) Feb 7, 2024
Magnificent 7 Total Returns over Last 3 Years... Nvidia $NVDA: +400% Microsoft $MSFT: +72% Meta $META: +71% Apple $AAPL: +40% Google $GOOGL: +40% Amazon $AMZN: +2% Tesla $TSLA: -35%
— Charlie Bilello (@charliebilello) Feb 7, 2024
Retrieval Augmented Generation (RAG) is emerging as the preferred technique for bringing context to LLMs We're launching to a new YouTube series to highlight RAG. Will start with the basics, but quickly move onto the advanced concepts (seems like theres a new one each week!) https://t.co/JCtHwesL7j
— Harrison Chase (@hwchase17) Feb 6, 2024
Meta presents Diffusion World Model DWM significantly surpasses one-step dynamics models with a 44% performance gain, and achieves SotA performance https://t.co/COCzD8krbr https://t.co/K6jahg0o2E
— Aran Komatsuzaki (@arankomatsuzaki) Feb 7, 2024
I just read Napoleon and Modern War. Here’s a list of ideas from the book: 1. Hesitation and half measures lose all in war. 2. Ten people who yell make more noise than ten thousand who keep silent. 3. A rapid march augments the morale of an army, and increases its means of… https://t.co/FOE4o2pxa1 https://t.co/2LlOcbQixj
— David Senra (@FoundersPodcast) Feb 6, 2024
Steve Jobs: "The vision needs to be reiterated." https://t.co/nVx8YClyFn
— Z Fellows (@ZFellows_) Feb 6, 2024
What is gRPC, and when should you use it? gRPC is a powerful remote procedure call (RPC) framework developed by Google, enabling efficient and fast communication between services. To understand gRPC, remote procedure calls must first be understood. A local procedure call is a… https://t.co/JXYhfDXAXX https://t.co/Rv5aUmjCcs
— Chris Staudinger (@ChrisStaud) Feb 6, 2024
You likely missed it if you only follow ML Twitter but there's a series of mind-blowing tech reports and open-source models coming from China (DeepSeek, MiniCPM, UltraFeedback...) with so much lesson learned and experiments openly shared together with models, data, etc This… https://t.co/tTvhm5c1Rl
— Thomas Wolf (@Thom_Wolf) Feb 6, 2024
🚀 𝐄𝐧𝐡𝐚𝐧𝐜𝐢𝐧𝐠 𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐰𝐢𝐭𝐡 𝐀𝐥𝐩𝐡𝐚 𝐓𝐮𝐧𝐢𝐧𝐠 𝐢𝐧 𝐇𝐲𝐛𝐫𝐢𝐝 𝐒𝐞𝐚𝐫𝐜𝐡 𝐢𝐧 𝐑𝐀𝐆 𝐰𝐢𝐭𝐡 @llama_index🌟 🤔 Ever wondered if your RAG application is effective for all kinds of user queries? For different user queries, a… https://t.co/YqcMaSNV8v https://t.co/PLQKx0SKvz
— Ravi Theja (@ravithejads) Feb 6, 2024
RAG From Scratch: Video series focused on understanding the RAG landscape RAG is central for LLM application development, connecting LLMs to external data sources. But, the pace of innovation and new approaches makes it challenging to keep up. We're launching a new video… https://t.co/Y9Ra26ZTE5 https://t.co/963lOnVLcP
— LangChain (@LangChainAI) Feb 6, 2024
Meta didn't release the text-based prompting capability of Segment Anything (SAM)😔 Luckily we can use @Google's OWLv2 to get boxes based on text, which we feed to SAM :) @mervenoyann created a nice Space for it here: https://t.co/4ienxEWSkJ https://t.co/N1clBOc80D
— Niels Rogge (@NielsRogge) Feb 5, 2024
It's obvious that generative AI will fundamentally reinvent marketing. Some of this reinvention is obvious: Just imagine content creation costs going to zero, as AI is certain to do. What happens next? Here are some ideas: - every individual marketer/small business will be able… https://t.co/6SPL7SmxlE
— andrew chen (@andrewchen) Feb 5, 2024
𝗪𝗵𝘆 𝗱𝗼𝗲𝘀 𝗚𝗼𝗼𝗴𝗹𝗲 𝗿𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱 𝗠𝗼𝗱𝘂𝗹𝗮𝗿 𝗠𝗼𝗻𝗼𝗹𝗶𝘁𝗵𝘀 𝗶𝗻𝘀𝘁𝗲𝗮𝗱 𝗼𝗳 𝗠𝗶𝗰𝗿𝗼𝘀𝗲𝗿𝘃𝗶𝗰𝗲𝘀? In the last decade, we have seen a massive trend of using microservices everywhere. We were building systems for a few hundred or thousand users and… https://t.co/ysuFpkE2MU https://t.co/8AvKTIqaLE
— Dr Milan Milanović (@milan_milanovic) Feb 5, 2024
We took apart the Vision Pro, and it is a phenomenal feat of engineering! There is just so much packed in this headset. Some highlights: https://t.co/m4I5LR6uoU
— Kyle Wiens (@kwiens) Feb 4, 2024
A lot of people who want to understand what’s different about ColBERT — but also want something shorter and simpler than reading the papers. This is it: https://t.co/qMgWdwGuK1
— Omar Khattab (@lateinteraction) Feb 4, 2024
Exclusive: Apple’s Phil Schiller says alternative app stores expose iPhone users to major risks. He’s right // this is correct. 1/2 https://t.co/BkujmYmdj4
— Steven Sinofsky (@stevesi) Feb 4, 2024
When I was first exposed to the Confusion Matrix, I was lost. And there was a HUGE mistake I was making with False Negatives. It took me 5 years to fix it. I'll teach you in 5 minutes. Let's dive in. 1. A confusion matrix is a tool often used in machine learning to visualize the… https://t.co/XA3LCqBLGs https://t.co/vy2IQDtorF
— 🔥 Matt Dancho (Business Science) 🔥 (@mdancho84) Feb 3, 2024
🚀 𝐂𝐨𝐫𝐫𝐞𝐜𝐭𝐢𝐯𝐞 𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥 𝐀𝐮𝐠𝐦𝐞𝐧𝐭𝐞𝐝 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧 (𝐂𝐑𝐀𝐆) ⭐ 🤔 RAG heavily relies on the relevance of retrieved-context, which can sometimes be fully correct or only partially correct, or incorrect. CRAG (Corrective Retrieval Augmented… https://t.co/sHn23417sJ https://t.co/pYeueGG0Yz
— Ravi Theja (@ravithejads) Feb 4, 2024
Peter Thiel on the importance of having a definite view of the future “We’re always focused on very short time horizons because you have to figure out a way to get through the next month, the next quarter, how you get some customers. But it’s always worth thinking ahead 5-10… https://t.co/YV4rxCNkZQ https://t.co/RTERMBWhYz
— Startup Archive (@StartupArchive_) Feb 4, 2024
Mobile-Agent Presents Mobile-Agent, an autonomous multi-modal mobile device agent. Leverages visual perception tools to perceive vision context and autonomously plan and decompose complex tasks and navigate mobile apps. The vision-centric gives it greater adaptability across… https://t.co/JG0sN70xGK https://t.co/MBTKFRt7uw
— elvis (@omarsar0) Feb 3, 2024
"To play a wrong note is insignificant. To play without passion is inexcusable." - Beethoven
— Lex Fridman (@lexfridman) Jan 8, 2024
12 RAG Pain Points and Proposed Solutions 💡 Building production RAG is hard. @wenqi_glantz compiled a list of 12 (!!) RAG pain points + added a full solution list to each one with @llama_index abstractions 🔥 We’ve put out cheatsheets before, but this one is much more… https://t.co/dnQ9jpIer8 https://t.co/dqBJC4c0yy
— LlamaIndex 🦙 (@llama_index) Feb 3, 2024
Netflix Tech Stack (CI/CD Pipeline) Let’s explore the innovative tools and techniques behind Netflix's world-class continuous delivery pipeline. Planning: Netflix Engineering uses JIRA for project planning and Confluence for documentation. Coding: Java is the primary language… https://t.co/nKVxUFdim5 https://t.co/NqeiWEgJlb
— Sahn Lam (@sahnlam) Feb 4, 2024
1/n Introducing RAPTOR Existing RAG methods suffer from a major limitation: they can only retrieve short, contiguous passages of text. This restricts their capacity to represent cross-document discourse structure and leverage thematic information scattered across lengthy… https://t.co/fZjhOs0F0q https://t.co/ZFBbb6yflG
— Carlos E. Perez (@IntuitMachine) Feb 1, 2024
