RAG parameters are not one-size-fits-all, and you should tune it! @ravithejads created a comprehensive walkthrough on how you should tune your retrieval parameters (specifically around hybrid search) based on properties of your query dataset. For some queries, having a low ฮฑ… https://t.co/ZeOl3zth1q https://t.co/J4yetCRpme https://t.co/BBf3oD6A1F
— Jerry Liu (@jerryjliu0) Jan 31, 2024
What are the most common ๐จ๐๐ฒ ๐๐ฎ๐๐ฒ๐ ๐ณ๐ผ๐ฟ ๐๐ฎ๐ณ๐ธ๐ฎ? We have covered lots of concepts around Kafka already. But what are the most common use cases for The System that you are very likely to run into as a Data Engineer? ๐๐ฒ๐’๐ ๐๐ฎ๐ธ๐ฒ ๐ฎ ๐ฐ๐น๐ผ๐๐ฒ๐ฟ ๐น๐ผ๐ผ๐ธ:… https://t.co/FpptkmIp99 https://t.co/wVqKsncl2K
— Aurimas Griciลซnas (@Aurimas_Gr) Jan 29, 2024
What idea changed how you view the world? Here's mine: High Agency. The high agency library: (19 best examples I've found in 4 years) https://t.co/Q6mviyMrvG https://t.co/IOttWTOC5u
— George Mack (@george__mack) Jan 28, 2024
Feynman's reply to a high school student who wrote to him asking for advice on how to prepare for college and a career in physics https://t.co/Bms8qAK3vl
— Fermat's Library (@fermatslibrary) Jan 27, 2024
Apple's proposed App Store and iOS changes seek to retain as much control over iOS as possible while staying within the letter of the EU's DMA (@gruber / Daring Fireball) https://t.co/Pyaqg2MmRI ๐ซ Subscribe: https://t.co/OyWeKSRpIM https://t.co/5fnFz32Zpt
— Techmeme (@Techmeme) Jan 27, 2024
weekend read: RAG vs Fine-tuning (agriculture case study) https://t.co/Z3wplB3ct7
— Suzana Iliฤ (@suzatweet) Jan 27, 2024
Decoder-only models only work with discrete tokens, right? ๐ค Excited to present ๐GIVT: Generative Infinite-Vocabulary Transformers, a simple way to generate arbitrary vector sequences with real-valued entries using transformer decoder-only models! https://t.co/ep2Ii2xXAP 1/ https://t.co/UexJdYCRMr
— Michael Tschannen (@mtschannen) Dec 5, 2023
๐ Complete Architecture for Building Enterprise RAG ๐ ️ This architecture by @rungalileo provides a reference on both the algorithmic and system-level components needed to build production RAG ๐ซ. There’s a lot in here ๐: It covers everything from the following broad… https://t.co/t5mIlf8ssU https://t.co/3zJ057GM86
— LlamaIndex ๐ฆ (@llama_index) Jan 27, 2024
You are invited! Join us in the @MSFTResearch Forum, a new quarterly series for sharing the latest ideas about research in the era of general AI. Explore and discuss the latest advances and ideas with the global research community. Registration is free. https://t.co/m6FrV36scA
— Peter Lee (@peteratmsr) Jan 26, 2024
Comparing ▪️ Prompt engineering ▪️ Retrieval-Augmented Generation (RAG) ▪️ Fine-tuning Here is what you need to know: https://t.co/Dfvo92H47v https://t.co/HLNIovvmB5
— TuringPost (@TheTuringPost) Jan 27, 2024
We’re excited to partner with @ExaAILabs - they’ve created the most advanced RAG-powered web search we’ve seen so far. ๐๐ฅ Unlike web search for humans, @ExaAILabs is tailor-made for LLMs, returning the most relevant highlights through custom chunking / extraction / retrieval.… https://t.co/1vZ79oobFk https://t.co/HXLMBMJZKc https://t.co/Ptqe8zHEek
— LlamaIndex ๐ฆ (@llama_index) Jan 26, 2024
I recently read Same as Ever by @MorganHousel. It was a phenomenal read packed full of timeless wisdom. Here are 25 of my favorite quotes from the book: https://t.co/F3UBYnRWJS
— Clay Finck (@Clay_Finck) Jan 26, 2024
Paul Graham: "One sign that determination matters more than talent: there are lots of talented people who never achieve anything, but not that many determined people who don't." https://t.co/AKmyhMePsu
— Brian Ji (@BrianJJi) Jan 25, 2024
The "Dense X Retriever" paper shows that it significantly outperforms the traditional chunk-based retriever @LoganMarkewich created an awesome LlamaPack that lets you get started with this proposition-based retriever in no time using @llama_index ๐ฅ Let's see how it works ๐๐งต https://t.co/f76qOk7lqe
— Rohan (@clusteredbytes) Jan 25, 2024
๐จ Breaking news: Google just launched LUMIERE, and it's crazy. This is going to transform the future of VIDEO forever Here are 5 stunning examples you can't miss out on: ๐งต https://t.co/m6jx86jqEl
— Shruti Mishra (@heyshrutimishra) Jan 25, 2024
I spoke with Clearbit on how they use AI. They were acquired for $150M by Hubspot last year. Here's what stood out from our conversation: - exploring use cases with an AI champion in charge - rolling it out to the team by forming small, cross-functional teams and speaking… https://t.co/gzYPrgOj41 https://t.co/OLlNZDBLAH
— Ben Tossell (@bentossell) Jan 25, 2024
NEWS: Porsche has unveiled their first ever all-electric SUV, the Macan EV. • Starting price: $80k ($107k for turbo) • Range: Up to 613km (380 miles), WLTP • 0-60mph in as little as 3.1s • 100 kWh battery • 800v • Rear wheel steering • Deliveries start late 2024 https://t.co/Eo0Yqpv8O0
— Sawyer Merritt (@SawyerMerritt) Jan 25, 2024
Elon Wuz Right: Vision is All You Need > TikTok team trains Depth Anything > LiDAR quality depth estimation from single photo frame > Using teacher model-student model system > 1st author Lihe Yang did this while on his internship (!) with the company Out of training set… https://t.co/NAl5dhzVOi https://t.co/IRqjZ335h6
— Ate-a-Pi (@8teAPi) Jan 25, 2024
Can LLMs themselves do retrieval/reranking? One of the main challenges includes being able to stuff enough document chunks into the LLM context window so it can decide which ones are most relevant. Had a short talk at the @ollama meetup last night (thanks @jmorgan and co. for… https://t.co/pVrlnvVOx0 https://t.co/c5z127PPZK
— Jerry Liu (@jerryjliu0) Jan 24, 2024
“All of us need to be on guard against arrogance, which knocks at the door whenever you’re successful.” ~ Steve Jobs https://t.co/Au2y9rdFTh
— Jon Erlichman (@JonErlichman) Jan 24, 2024
What are the main ๐๐ฒ๐ฒ๐ฑ๐ฏ๐ฎ๐ฐ๐ธ ๐๐ผ๐ผ๐ฝ๐ ๐ถ๐ป ๐ ๐ฎ๐ฐ๐ต๐ถ๐ป๐ฒ ๐๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด ๐ฆ๐๐๐๐ฒ๐บ๐? A regular Machine Learning Project consists of 4 main stages: ๐ญ: Ideation stage where business meets data professionals to raise a hypothesis that is then evaluated against the… https://t.co/XFS8wVI5nd https://t.co/sgSSGAuGYc
— Aurimas Griciลซnas (@Aurimas_Gr) Jan 23, 2024
Basic chat is easy, but a delightful chat experience has a lot of subtleties! In today's OSS project spotlight, https://t.co/COUh9LAKzR from @charlespacker is an advanced chat bot built with LlamaIndex. MemGPT uses function-calling by LLMs under the hood to maintain multiple… https://t.co/Zd0d9Aczk3 https://t.co/iZ0mRQUBbq
— LlamaIndex ๐ฆ (@llama_index) Jan 23, 2024
๐ฅ๐ฅ ๐ฅMulti Agent system monitor and answer my emails! If you have a busy inbox, work at sales or anything like that, this one is for you! Using CrewAI ๐ฃ and LangGraph ๐ธ️ - You can check the video: https://t.co/Q2WrJTNYMG - Code is all opensource: https://t.co/ayb2QSSZLr https://t.co/SNm3znpvxj
— Joรฃo Moura (@joaomdmoura) Jan 23, 2024
This $25M+ business was validated in 3 hours for $5: Problem: You’re stuck in NYC with all your bags, but no hotel Solution: Bag storage as a service. How the founder validates: 1. DOES NOT write a single line of code 2. Creates a BASIC landing page w/ Typeform (see image) 3.… https://t.co/pAKUv33tkZ https://t.co/uVHbyUF3yR
— Pat Walls (@thepatwalls) Jan 23, 2024
Building the LLM OS requires memory, and memory is a much deeper problem than top-k RAG. MemGPT is a cool paper by @charlespacker that takes inspiration from hierarchical memory in traditional OS’s, to automatically page “in” and “out” information from an underlying storage… https://t.co/HG3YBNcynP https://t.co/8g9EfyEbHK https://t.co/v9NDBLGgYo
— Jerry Liu (@jerryjliu0) Jan 24, 2024
ML YouTube Courses (~12K⭐️) It's impressive how many high-quality free AI courses there are. To make them easy to find, I maintain this collection with some of my favorite courses around topics like NLP, deep learning, LLMOps, maths for ML, and more. This has been useful for… https://t.co/M6g76X5i0y https://t.co/XYX6zUMKz1
— elvis (@omarsar0) Jan 23, 2024
Knowledge Fusion of LLMs Is it possible to merge existing models into a more potent model? We have already seen a few ways that show the potential to effectively do this using approaches like weight merging and ensembling of models. This work proposes FuseLLM with the core… https://t.co/EwTAChDY7A https://t.co/xu5E86zIo4
— elvis (@omarsar0) Jan 22, 2024
Real time deep fakes running locally on M1 Pro. Im scared...๐ณ Using https://t.co/TSed4dIkE7 On https://t.co/nSuzWGgcsb https://t.co/epWYyBcgQ3
— nico-lรกs (@nos_ult) Jan 21, 2024
Interesting https://t.co/meQ12RyYBP
— DreamsHorizon (@DreamsHor1zon) Jan 20, 2024
Apple Ferret, a Multimodal LLM capable of understanding spatial referring of any shape or granularity within an image and accurately grounding open-vocabulary descriptions. I hope this will use Apple MLX for training soon ๐ https://t.co/RM8z5AK0Lk https://t.co/a5R32YlK4d
— ifioravanti (@ivanfioravanti) Jan 21, 2024
7/ Overview of LLMs for Evaluation - thoroughly surveys the methodologies and explores their strengths and limitations; provides a taxonomy of different approaches involving prompt engineering or calibrating open-source LLMs for evaluation. https://t.co/iMKw3mAheT
— DAIR.AI (@dair_ai) Jan 21, 2024
3/ RAG vs. Finetuning - report discussing the tradeoff between RAG and fine-tuning when using LLMs like Llama 2 and GPT-4; performs a detailed analysis and highlights insights when applying the pipelines on an agricultural dataset. https://t.co/9y2aycA5e3
— DAIR.AI (@dair_ai) Jan 21, 2024
The Top ML Papers of the Week (Jan 15 - Jan 21): - AlphaCodium - AlphaGeometry - RAG vs. Finetuning - Self-Rewarding Models - Overview of LLMs for Evaluation - Tuning Language Models by Proxy ...
— DAIR.AI (@dair_ai) Jan 21, 2024
Looks like 2024 is the year when we ๐to๐ in the text-to-image and text-to-video domain! @EMostaque https://t.co/6NM94nUuyz https://t.co/Y2QqMGtW5k
— Gradio (@Gradio) Jan 20, 2024
What on EARTH is Chick Fil A doing right and how did Subway, Arbys, BK and KFC get it so wrong? 2022 QSR Average Unit Sales Chick Fil A: $6.7 m Raising Canes: $5.4m Shake Shack: $3.8m Whataburger: $3.7m McDonalds: $3.6m Culvers: $3.3m Panera Bread: $3.2m Chipotle: $2.8m Krispy… https://t.co/tbGu9gTBgA
— Triple Net Investor (@TripleNetInvest) Jan 21, 2024
Your brain is a voxel. https://t.co/2eJcjddA3g
— Robert Scoble (@Scobleizer) Jan 20, 2024
This paper starts out amazing and just keeps getting better https://t.co/Y7qKI0CKD5 https://t.co/7f9G6QbdBu
— Jonty Wareing ⍼ (@jonty) Jan 20, 2024
Enjoying the style of MetaAI's papers. They're almost complete recipes (ingredients not included) with valuable ablations, under 10 pages. • Search: https://t.co/zxPKtrTFyv • DPR: https://t.co/hKzFFziMtF • Humpback: https://t.co/a8EHiFFgJp • Llama: https://t.co/TRzWY0aKjJ,… https://t.co/0Cl7jF46Iq
— Eugene Yan (@eugeneyan) Jan 20, 2024
Control the lighting of your Midjourney creations Here is a list of setup that will most definitely yield you some fantastic results: — Backlighting — Soft overhead light — Lens flare burst — Rim lighting — Underexposed silhouette — Rembrandt lighting — Dappled Sunlight — Halo… https://t.co/1d3R7bKt3O https://t.co/hWlwB3FTtq
— Hugo Ventura (@hugovntr) Jan 20, 2024
Fallingwater, the modern house prototype built by Frank Lloyd Wright over a waterfall in 1935 https://t.co/1C0JclfWcT
— Massimo (@Rainmaker1973) Jan 19, 2024
"RAG in a notebook" is usually over a static data source. A step towards production is batch updates (e.g. every hour). Live production RAG is giving it real-time access to all information updates in your knowledge corpus. Here's a great reference architecture by @kiennt_ on an… https://t.co/0PGrOklPQn https://t.co/c4AlRqYi93 https://t.co/eWEjqtjFLU
— Jerry Liu (@jerryjliu0) Jan 20, 2024
๐จ EXCELLENT PAPER WARNING: check out "Generative AI in EU Law: Liability, Privacy, Intellectual Property, and Cybersecurity." Important info and quotes below: The paper was published 2 days ago by @fecasolari, @philipphacker15, @GSpedicato & @Floridi, and it covers (some of my… https://t.co/vrwcpyRHy2 https://t.co/JpC3D0ffTy
— Luiza Jarovsky (@LuizaJarovsky) Jan 19, 2024
So the UK govt has just dropped a 'generative AI framework' for public service delivery. A few interesting things buried in this which may hint at the direction of UK from a policy/delivery side 1/5 ๐งต https://t.co/nUVdCSA4KX
— Tom Westgarth (@Tom_Westgarth15) Jan 19, 2024
Software 3.0 https://t.co/tTSlIvlMhv https://t.co/IAaaoMgEok
— Carlos E. Perez (@IntuitMachine) Jan 19, 2024
This is why Steve Jobs was a marketing genius: https://t.co/hEdjxjLakI
— MATT GRAY (@matt_gray_) Jan 19, 2024
Multi-Stock Ticker Analysis in One Line of Code ๐๐ Our new LlamaPack by anoopshrma is a great finance starter kit. Define any number of stock tickers, the time range you want to look at, and ask any structured question over any stock ticker. Possible with our… https://t.co/6ZtVRgKhGm https://t.co/zSqoJq78Ps
— LlamaIndex ๐ฆ (@llama_index) Jan 19, 2024
Good pipelines will always be key. That was true for DNN layers, now true for LMs. Other lessons from DNNs: finding reuseable modules (layers with parameters) and optimizers! If only we had a concrete framework with control flow design patterns and optimizers for LM programs๐ https://t.co/lZ2vvKDT1B https://t.co/TXhmgv05eO
— Omar Khattab (@lateinteraction) Jan 19, 2024
Introducing RowsX: Scrape any table or list from any website effortlessly. Converts web data into spreadsheets with a single click. RowsX can extract data from tables, product listings, articles, and search results. Check it out here: https://t.co/xx1SXixOqh https://t.co/dLRX6AYVry
— Hasan Toor ✪ (@hasantoxr) Jan 18, 2024
Advanced Query Transformations to Improve RAG ๐ A good way improve the reasoning capability of RAG is to add a query understanding layer - add query transformations like HyDE, sub-question decomposition, iterative reasoning, or routing before actually querying the vector store.… https://t.co/bqztBSxXOo https://t.co/iiLsS7uhNX
— LlamaIndex ๐ฆ (@llama_index) Jan 19, 2024
I spoke to @andre_z from @qdrant_engine , @jerryjliu0 from @llama_index, & @Nils_Reimers from @cohere about the exciting and untapped potential of embedding models beyond document retrieval and RAG. Expect exciting developments in 2024! https://t.co/RW6FqPs6IN
— Ben Dickson (@bendee983) Jan 18, 2024
Overview of LLMs for Evaluation LLMs as tools for evaluation is growing rapidly. This new paper thoroughly surveys the methodologies and explores their strengths and limitations. There are a lot of creative ways in which LLMs are used for evaluation. The taxonomy below… https://t.co/AjNtbVzhwy https://t.co/R1D8j01NR6
— elvis (@omarsar0) Jan 18, 2024
A lot of advanced RAG techniques involves expressing your retrieval/query pipeline as a hierarchy: get some initial set of “nodes”, retrieve more nodes based on the initial set, and so on. This includes multi-document use cases, complex documents, and multi-modal settings. Our… https://t.co/Rj5vLoty2Z https://t.co/VrXGsmLFL7
— Jerry Liu (@jerryjliu0) Jan 18, 2024
Foundations of Vector Retrieval Very rare drop on arXiv. This 200+ page monograph covers fundamental concepts along with advanced data structures and algorithms for vector retrieval. Given how much we deal with vector representations today and how critical they are to… https://t.co/BIVCuMrJHp https://t.co/YxnrYJtBSk
— elvis (@omarsar0) Jan 18, 2024
Global chip revenue dropped 11% YoY to $533B in 2023; Intel regained top spot with $48.7B, followed by Samsung's $39.9B; Nvidia was top five for the first time (Gartner) https://t.co/lMs3JyzMFk ๐ซ Subscribe: https://t.co/OyWeKSRpIM https://t.co/fpPbWRh3bN
— Techmeme (@Techmeme) Jan 17, 2024
RAG vs. Fine-Tuning Cool report discussing the tradeoff between RAG and fine-tuning when using LLMs like Llama 2 and GPT-4. It performs a detailed analysis and highlights insights when applying the pipelines on an agricultural dataset. Here is a figure showing the pipeline… https://t.co/ZlOvpgre68 https://t.co/G3r6G8nzkc
— elvis (@omarsar0) Jan 17, 2024
Every country needs an AI strategy. Good to see oen for Japan here and for India in @SarvamAI https://t.co/rmUIZlS6Fd
— Vinod Khosla (@vkhosla) Jan 17, 2024
“The Expressive Power of Transformers with Chain of Thought” w/ @Ashish_S_AI will appear at ICLR ๐ฆ๐น https://t.co/ay76PikxY0
— Will Merrill (@lambdaviking) Jan 16, 2024
We’ve talked a lot about advanced RAG, but question-answering over tabular data is just as important, and our new cookbooks show you how to build advanced tabular QA from scratch ๐ซ Learn how to build a Pandas or SQL-powered query engine that goes beyond simple prompting - learn… https://t.co/n0BgrTkN4a https://t.co/U40bsDs4f7 https://t.co/qBIfkkx9kP
— Jerry Liu (@jerryjliu0) Jan 16, 2024
When I was raising money for my startup, I had a hard time knowing which VC funds I should be talking to. I wanted to see a list of all VC funds that invest in seed stage startups. I found this list the other day of all VC funds under $200m. and now I'm sharing it.
— Ryan Shank (@ryanashank) Jan 14, 2024
How RAGs (Retrieval Augmented Generation) work... #infographic source: https://t.co/PTD1HkhS8i #GenerativeAI #VectorDB #AI #MachineLearning #DeepLearning #DataScience #Knowledgebase #LLMs https://t.co/uN42hbTFZR
— Kirk Borne (@KirkDBorne) Jan 15, 2024
Is synthetic data the future of AI? Introducing WebSight - a multimodal dataset featuring 823,000 pairs of synthetically generated HTML/CSS codes along with screenshots of the corresponding rendered websites, to fine-tune GPT4-V-like models. ๐๐ป Details in the thread ⬇️ https://t.co/xSWQIQ5Tgu
— Hugo Laurenรงon (@HugoLaurencon) Jan 15, 2024
Create your very own AI Influencer with Midjourney v6 ! Despite not being the "go-to" AI to do it, Midjourney is pretty capable when it comes to using images as references. But the easiness of execution is where it shines! Here is the 2 step process: 1. Generate your reference… https://t.co/43rBYnyG45 https://t.co/fBurd9xrS5
— Hugo Ventura (@hugovntr) Jan 15, 2024
The premium AI tools continue to become a competitive landscape. https://t.co/gWe0Wb8JGe
— Ben Bajarin (@BenBajarin) Jan 15, 2024
Gartner predicts that by 2030, machine learning models will use synthetic data exclusively, making real data unnecessary. Here are 10 companies you can use to generate synthetic data: @gretel_ai: Offers data anonymization solutions with APIs and machine learning models for data… https://t.co/uxJRgb05oW https://t.co/Q1A90C8fhQ
— TuringPost (@TheTuringPost) Jan 15, 2024
An unsettling listen, because Ovitz goes into vivid detail about just how much work building a career like his takes. It's not just talent, creativity, charisma, rigor, and consistency- it's sheer hours. I was trying to crunch it and it seems like it was 90-100 hour weeks for a… https://t.co/r6zmabnCBJ https://t.co/oiUsYngFo5
— Hiten Samtani (@hitsamty) Jan 14, 2024
Thoughts on the new Anthropic paper: (1) It's seminal, really useful, and a slam dunk in general. (2) But it's not the *first* evidence we have of insidious failure modes evading adversarial training in LLMs. I compiled some related work here: https://t.co/M2w7l9LCo5
— Cas (Stephen Casper) (@StephenLCasper) Jan 14, 2024
Idk if people noticed but Mixtral-Instruct was trained with Direct Preference Optimization (DPO) My prediction that a DPO variant will replace RLHF is already coming true https://t.co/tLDIeolSOa
— Nora Belrose (@norabelrose) Jan 14, 2024
The cloud chamber was invented by physicist Charles Wilson, who was awarded the Nobel Prize in Physics in 1927 for his invention. His invention was a major breakthrough in particle physics as it allowed for the visualization and study of ionizing radiation and subatomic… https://t.co/qbcnXh2lXR https://t.co/sfjfjrdPKf
— Fermat's Library (@fermatslibrary) Jan 14, 2024
Here’s a neat paper by Barnett et al. (@DeakinA2I2) that outlines 7 failure points in building a RAG pipeline over your data. ๐ซ Missing content (did not index it) ๐ซ Missing in top-k retrieved set ๐ซ Missing in reranked set ๐ซ Not extracted (in context but LLM couldn’t use) ๐ซ… https://t.co/UvSRBxDL2n https://t.co/nftC9aZ6Jv
— Jerry Liu (@jerryjliu0) Jan 14, 2024
According to Bloomberg Intelligence, the GenAI market is projected to rise from $40 billion in 2022 to $1.3 trillion by 2032. It's a compound annual growth rate (CAGR) of 42%! Here are the main conclusions from this analysis: https://t.co/xneTegwUUI
— Ksenia Se (@Kseniase_) Jan 13, 2024
Hey, I'm ex-Reddit advisor and sold a community platform to WeWork. I noticed something BIG recently if you're looking to build cash-flowing internet businesses: There are 1.8B active users in Facebook Groups in 2024 But I've noticed 95% of paid communities are dead. RIP.… https://t.co/JeRNsAwRoU https://t.co/mSah5mJpdz
— GREG ISENBERG (@gregisenberg) Jan 13, 2024
Apptopia: WhatsApp's daily business users increased by 80% in the US in 2023 after a 122% jump in 2022, and daily users in the US grew by 9% in 2023 (@kantrowitz / Big Technology) https://t.co/lgpNiKAKtY ๐ซ Subscribe: https://t.co/OyWeKSRpIM https://t.co/o55XMmIQms
— Techmeme (@Techmeme) Jan 12, 2024
Undoubtedly, AI is the future of the tech landscape. I sat down for a Q&A to share thoughts on the emergence of this innovative tool and how our team at Blue Owl Capital views the benefits and possible drawbacks of artificial intelligence. Follow for more insights. https://t.co/F6E6MzKy1X
— Marc Lipschultz (@MarcLipschultz) Jan 10, 2024
Mark Zuckerberg shares the best advice Peter Thiel ever gave him "In a world that's changing so quickly, the biggest risk you can take is not taking any risk." Whenever you're in a position where you have to make a big shift in the direction of the company or product, people… https://t.co/qHgKL0jI3B https://t.co/Qo7Ckpv7KC
— Startup Archive (@StartupArchive_) Jan 13, 2024
i'm so ready for the education, entertainment AI revolution ๐คฃ๐ https://t.co/Zwib1DLxXz
— afterveil (@afterveil) Jan 13, 2024
There will never be a better time to build and rank a GPT than today. The current top-ranking GPTs have a major first-mover advantage based on the setup of the GPT store. Here's the best guides on how to build a high-value GPT to get you started:
— Rowan Cheung (@rowancheung) Jan 13, 2024
Building Production-RAG (systems edition!) Besides accuracy, speed is also important. Here's some tips on creating an efficient RAG pipeline over a large amount of data ⚙️ ๐ Some tips are library-specific, some are infrastructure ๐ 1. Implement parallel… https://t.co/8urlkvdg1S https://t.co/g9NT4ro9WZ
— Jerry Liu (@jerryjliu0) Jan 13, 2024
This is amazing. https://t.co/mtZwmq9kDK
— Jesse Proudman (@jesseproudman) Jan 12, 2024
๐จ#PromptShare๐จ Let's dive into a divine #prompt: highlight your favorite gods or anything you're inspired by in this theme. [character/god] of white color flying through the clouds, intricately sculpted, solarization effect, mixes realistic and fantastical elements,… https://t.co/AjNgbZkA1r https://t.co/b7bGZUebzg
— Pierrick Chevallier | IA (@CharaspowerAI) Jan 12, 2024
Agentic Chunking? We should chunk our text off its meaning...not physical characters Use an agent-like app to look a text, have it determine if it should be joined with a chunk, or create a new one Yes it's slow and expensive, but most solid and tunable results I've seen https://t.co/fpMa3t6LKs
— Greg Kamradt (@GregKamradt) Jan 12, 2024
What Generative AI Reveals About the Human Mind, by Andy Clark https://t.co/SIhQQAmFTw ... I read this as arguing that real AGI will require agency
— Kevin Mitchell (@WiringTheBrain) Jan 12, 2024
The new Together Embeddings ๐ซ has some pretty nice properties: 1. SOTA retrieval quality 2. 32k context length (embed big documents. My take is you probably want different embeddings at different levels of granularity, to better control for retrieval accuracy and cost. But this… https://t.co/M7Caaze86f https://t.co/U9kW2Mxi8t https://t.co/ZWiYEnxpju
— Jerry Liu (@jerryjliu0) Jan 12, 2024
Towards Conversational Diagnostic AI abs: https://t.co/HmYK2vchuh Another amazing paper from the Med-PaLM team at @GoogleDeepMind! Introduces AIMIE, an LLM based AI system optimized for clinical history-taking and diagnostic dialogue. Trained on medical QA, summarization, and… https://t.co/Gsf74Vsx7n https://t.co/beuumGOWrE
— Tanishq Mathew Abraham, Ph.D. (@iScienceLuvr) Jan 12, 2024
i like to think this is one of the key insights (that fits on the back of an envelope) that he was talking about https://t.co/uu2ZbujpY5 https://t.co/mONraL13Qr
— Yohei (@yoheinakajima) Jan 12, 2024
Predictions for AI in 2024 from Max Hjelm, VP of Sales at @CoreWeave 1. Will the GPU craze continue? 2. Computing and generative AI 3. Modern software stack ๐งต https://t.co/kN87cYzJUU
— TuringPost (@TheTuringPost) Jan 10, 2024
1-10-2024 ($) •AI at CES •The Rabbit R1 https://t.co/09HpA8bCta
— Stratechery (@stratechery) Jan 10, 2024
Fantastic to be at #CES2024 to talk about our longstanding partnership with @Walmart, and how they are applying the power of Azure OpenAI Service, along with their own proprietary data and models, to transform the digital shopping experience. https://t.co/28ff4tMBxs
— Satya Nadella (@satyanadella) Jan 9, 2024
The most interesting thing about the @LangChainAI v0.1 announcement is the release of `langgraph` (https://t.co/Evnh2iF6ak), which allows you to compose reasoning "workflows" in a graph-like fashion, where the agent is able to make decisions on which steps to take next. If… https://t.co/zxe3cnwgCn
— Michael Struwig ๐ฆ (@MichaelNStruwig) Jan 9, 2024
When I saw @MistralAI's pitch deck in June 2023, I didn't believe they would release an open GPT-3.5 level model by the end of the year. Fast forward 6 months and we have Mixtral-8x7B, a model that is on par with it/better according to @lmsysorg's Arena! https://t.co/nBEJDOPHxn https://t.co/spZO6JXJ88
— Niels Rogge (@NielsRogge) Jan 9, 2024
3,072 MI250X used to train 22 billion, 175 billion, and 1 trillion parameter LLM They showed tons, from optimizers to distributed parallelism strategies to hyperparameter tuning + search 32% to 38% flops utilization which is okay, but much lower than A100 https://t.co/kjCRUNdqZb https://t.co/zTz4Fq1cW7
— Dylan Patel (@dylan522p) Jan 9, 2024
Multilingual RAG is important! ๐ SeaLLM-Chat-7B is a cool model that is optimized for SEA languages: Vietnamese, Indonesian, Thai, Malay, Khmer, Lao, Tagalog, Burmese. Check out how to build a RAG pipeline over your Thai documents below ๐ Post: https://t.co/L2T09pZP6X… https://t.co/hU0Q8KP2Va https://t.co/uhPIhc9sAs
— Jerry Liu (@jerryjliu0) Jan 9, 2024
At @Gucci, Salesforce Service Cloud and EinsteinAI are the newest trusted team members making waves! ๐ค✨ Joining 600+ advisors, crafting replies from customer data, and upholding the iconic Gucci voice. ๐️ Plus, it's turbo-charging onboarding! https://t.co/fpP34rmnHb
— Marc Benioff (@Benioff) Jan 9, 2024
NEW PAPER: Mixtral of Experts ๐งกhttps://t.co/GqndXDa7HM Check out how Mixtral 8x7B works! We are also hosting an office hour on @MistralAI Discord tomorrow. Join us if you have any questions about Mistral models or la plateforme: https://t.co/Ba6nLn9h3D https://t.co/NiHBFhdn3r
— Sophia Yang, Ph.D. (@sophiamyang) Jan 9, 2024
How to choose a vision model for your specific needs? How do ConvNet / ViT, supervised / CLIP models compare with each other on metrics beyond ImageNet? Our work comprehensively compares common vision models on "non-standard" metrics. (1/n) https://t.co/LSXmEQiD2K
— Zhuang Liu (@liuzhuang1234) Jan 8, 2024
There’s some common LLM-powered workflows that users are orchestrating over their data: * Structured data extraction: input -> prompt --> LLM —> output parser * Naive RAG: input -> retriever -> synthesizer * Advanced RAG: input —> query rewriting —> retriever —> reranker —>… https://t.co/0bK0UTfn7B https://t.co/TedK7CSz11 https://t.co/Z6FO7Hs5M0
— Jerry Liu (@jerryjliu0) Jan 8, 2024
Quora founder Adam D’Angelo explains cohort retention and other startup metrics you should track In the clip below, the Quora founder and former CTO of Facebook argues: “As a startup, you really need to focus. So the core concept that I would try to focus on is users that are… https://t.co/tb5n4Jf7Ka https://t.co/tkSgSMr4hT
— Startup Archive (@StartupArchive_) Jan 8, 2024
This is such a fun Runway Workflow AND It creates a pretty seamless loop ๐คค 1. Construct a rough 3D environment with the basic composition of your scene. a. Position a camera to render an image from the perspective that you desire. b. Render Image (I'll share my input… https://t.co/ONnSmdjsLY https://t.co/AEiwfu20hd
— ๐ฅ Timmy ๐ฅ (@IXITimmyIXI) Jan 6, 2024
"Lee Kuan Yew took Singapore from Third World to First...If there is a lesson from Singapore’s development it is this: forget grand ideologies and others’ models. There is no replacement for experimentation, independent thought, and ruthless pragmatism." https://t.co/oceYIXBMiO
— Rob Henderson (@robkhenderson) Jan 6, 2024
A really incredible book. https://t.co/kh9LlDzdrY https://t.co/H8hLWm0XBh https://t.co/1HOUkDRxKB
— Marc Andreessen ๐บ๐ธ (@pmarca) Jan 7, 2024
I just got back from a 2-week trip this morning where I was visiting my family over the holidays, and I'm checking in to see what I've missed in January 2024. 5 minutes later, there were already +3 super interesting weekend reads on my desk. It's going to be an exciting year!… https://t.co/KWlw9QRmnC https://t.co/5eMO7gjZLA
— Sebastian Raschka (@rasbt) Jan 7, 2024
Sam Altman's 9 things that the best founders do to build a great company #1 Get to know your users really well “The best founders do customer support themselves. They go visit their users—in the case of Airbnb they go live with them. You want to get to know your users really… https://t.co/iwwXS22336 https://t.co/cdTjIMfR6g
— Startup Archive (@StartupArchive_) Jan 7, 2024
A curious criterion. https://t.co/6E9ZA8qb3P
— Carlos E. Perez (@IntuitMachine) Jan 7, 2024
If LLM composition is of interest, then you will find CALM relevant! Please check it out. https://t.co/EcaqJBjTYx
— Partha Talukdar (@partha_p_t) Jan 6, 2024
Advanced RAG Cheat Sheet + Recipes ๐ง๐ณ We’re publishing a comprehensive diagram outlining all the different components of advanced RAG, the pain points they solve, and links to LlamaIndex resources. Here’s some core concepts that motivate advanced RAGs: ๐ก Success metric for RAG… https://t.co/E3a44cmBU3 https://t.co/ypZ2YYaXg9
— LlamaIndex ๐ฆ (@llama_index) Jan 5, 2024
Extending an LLM for new knowledge sources is tedious—fine-tuning is expensive/causes forgetting, LoRA is restrictive. Excited to share our work where we show that an LLM can be efficiently *composed* with specialized (L)LMs to enable new tasks! https://t.co/DBZ8aQNQTo ๐งต(1/8) https://t.co/R0h4JEFid8
— Rachit Bansal (@rach_it_) Jan 5, 2024
Is Google Sheets the only LLMOps you need? I've used the same format for the last 6 years to track experiments. It works for speech, vision and text models, it’s that flexible. Here's what I found useful to track (sheet link below) - description: single sentence that… https://t.co/j3pfcxC9oX https://t.co/xuELxanftA
— Arnaud Benard (@arnaudai) Jan 3, 2024
What we'll see in 2024: • Models become a commodity; instruction data & reward functions are the moat • Model-generated data (RLAIF, REST^EM) increasingly outperforms human annotations • Model API costs plunge; we mostly finetune for legal, privacy, task performance reasons
— Eugene Yan (@eugeneyan) Jan 3, 2024
2024 DARPA & DoD budgeted project highlights: - multiple inferences of impending supply-chain collapse & bioengineered microbial nutrient solutions - multiple projects related to operating in extreme cold, electromagnetically unpredictable, austere & GPS-compromised environments https://t.co/paqzXilyRl
— eleven (@eleevn) Jan 3, 2024
My goal for 2023 was to hit $10k MRR https://t.co/aVdLRGEeK0 just surpassed $220K MRR The goal for this year is $1M MRR, here is how I will get there https://t.co/a2pRqYAHVU
— Yasser (@yasser_elsaid_) Jan 3, 2024
In 1978, James Burke made this perfectly timed shot on television and is widely considered "The Greatest Shot In Television History" https://t.co/g7lw8zgwjJ
— Fascinating (@fasc1nate) Jan 3, 2024
Fine-tune a Mistral-7b model using Direct Preference Optimization (DPO). Just published a tutorial on @TDataScience about using DPO to enhance the performance of SFT models. Funnily enough, I created NeuralHermes-2.5 for this article. https://t.co/XyrcXOZ0Ed
— Maxime Labonne (@maximelabonne) Jan 2, 2024
Some great notes on fine tuning Mistral 7B with MLX: “MLX is fast. This took 3:24 hours as opposed to almost 12 hours using llama.cpp” And its going to keep getting faster ๐ https://t.co/whpcRvJj3K https://t.co/Xr5ymtRyQP
— Awni Hannun (@awnihannun) Jan 2, 2024
Peter Thiel on what he would look for if he was joining a startup In the clip below, Wharton professor Adam Grant asks Peter Thiel what he would look for if he was joining an early-stage startup. Thiel gives a simple response: “Do you like the people? Do you think you could… https://t.co/Ao0cxRujIV https://t.co/AE9NZl6XvY
— Startup Archive (@StartupArchive_) Jan 2, 2024
Netflix just casually released a comprehensive bi-annual spreadsheet of hours watched for every title (over 50,000 hours)... The most watched show between January to June 2023 was: The Night Agent (Season 1) https://t.co/PHCWW6lYdt
— Blake Robbins (@blakeir) Jan 2, 2024
Here’s a compilation of my recent work and a sneak peak at what you can expect from me in 2024. All this was achieved in the last 6 months. Although I do sometimes work 18 hour days and often forget to eat. Thank you to @runwayml, @pika_labs, @midjourney, @elevenlabsio,… https://t.co/XloJ7L8LKv https://t.co/bhJZ1Yr8SD
— Jeff_synthesized (@JeffSynthesized) Jan 2, 2024
The most urgent person usually wins. —@karalawson20, head coach @DukeWBB https://t.co/hpSLiRHpSZ
— Vala Afshar (@ValaAfshar) Jan 2, 2024
Anyone that listens to @FoundersPodcast knows David does insane research (see picture of just one book and his highlights!) You can now access all of them at https://t.co/FxCq6F5NPP It’s more or less like that Matrix thing where he learns Kung Fu, but for founders https://t.co/FZiUjO6vsm
— Patrick OShaughnessy (@patrick_oshag) Jan 2, 2024
Wow. Here’s the the ultimate guide to AI-generated characters! ๐ by @vladerosh_ai Learn step-by-step how to create any consistent character, in any pose: a. Generate a training dataset with Scenario b. Train a custom model in minutes c. Prompt it to bring your vision to… https://t.co/QSjNgHEtRh https://t.co/f8EHnMJmn9
— Emm (@emmanuel_2m) Jan 2, 2024
How to launch an ecom business and do $10,000 per month. 1. Use https://t.co/MO3m4Chrd9 to find a product with negative Trust Pilot Reviews. 2. Study the negative reviews. 3. Find a similar supplier on Alibaba and improve the product. https://t.co/8vO1KhXqjB
— daviefogarty (@daviefogarty) Dec 10, 2023
The History of Windows Millennium (Windows Me) https://t.co/wi36Ybflmi // Often forgotten this post captures the relative chaos of the last "16-bit" version of Windows at an awkward transition time for Microsoft from consumer/retail to enterprise-focused.
— Steven Sinofsky (@stevesi) Jan 1, 2024
I think the business of religion is going to explode, because of Gen Z and America's loneliness epidemic. It already generates more revenue than APPL ($383B), GOOGL ($297B) and MSFT ($211B)…combined So deep dived and came up w 18 business ideas to capitalize on the trend: ๐งต https://t.co/hFgQcUrFay https://t.co/ptVkrUjv86
— AJ Chan (@mrajchan) Jan 1, 2024
Reading this today evening. 'How to do Great Work" - Paul Graham https://t.co/3tkVCbFZHu https://t.co/k89yeymfvK
— Vipul Vaibhaw (@vaibhaw_vipul) Jan 1, 2024
AI and the Kubler-Ross stages of grief curve. Denial Refusal to acknowledge the advancements in AI and cognition. Ignoring the implications of AI breakthroughs like GPT-4. Maintaining a belief in outdated cognitive theories. Anger Frustration with the disruption caused by AI… https://t.co/I9l8ouEAKM https://t.co/SNrbuKodFR
— Carlos E. Perez (@IntuitMachine) Jan 1, 2024
Tips for Amazon Writers (source: https://t.co/yxSFs9xOHP) https://t.co/HSDCCZbqEd
— George from ๐นprodmgmt.world (@nurijanian) Dec 31, 2023
