Today a deepfake img caused a $100B market swing. This could have been prevented by requiring @Twitter to tag AI-generated content. Here is a great @CNBC @SullyCNBC post featuring Ben Colman & @DetectDeepfakes https://t.co/CBBjIjTl2N
— Starson (@DrStarson) May 23, 2023
In the final chapter of software eating the world, traditional SaaS itself may be on the menu. Here's how the code-generation capabilities of generative AI could disrupt the way software is created, distributed, and used. https://t.co/DcfEz1a0Tt
— Stephanie Palazzolo (@steph_palazzolo) May 22, 2023
LIMA - a new 65B parameter LLaMa model fine-tuned on 1000 carefully curated prompts and responses. A few important takeaways from this work: - doesn't use RLHF - generalizes well to unseen tasks not in the training data - LIMA responses are equivalent or preferred to GPT-4 in… https://t.co/89HF3SJQnv https://t.co/ygLzmk0tZS
— elvis (@omarsar0) May 22, 2023
LIMA : LLaMA 65B + 1000 supervised samples = {GPT4, Bard} level performance. From @MetaAI https://t.co/FIuIo6agXa
— Yann LeCun (@ylecun) May 22, 2023
1/9 Another excellent Caixin article on the intractability of the local-government debt burden. In discussing the various forms and channels of hidden debt raised by local governments, it notes that "under pressure to meet growth and... https://t.co/pPBso6IoLx
— Michael Pettis (@michaelxpettis) May 22, 2023
LLM Pre-training vs. Instruction-Tuning -LLaMa 65B pre-trained -Only simple fine-tuning, w/ only 1k (carefully chosen) data points, no RLHF -Can plan trips & speculate about alternate histories -Generalizes to unseen tasks -Humans prefer it over GPT-3 https://t.co/6edcWjVz04 https://t.co/wQvbbxpmtc
— John Nay (@johnjnay) May 22, 2023
How Does Generative Retrieval Scale to Millions of Passages? Finds that the use of synthetic queries as a document representation strategy is the only approach that remained effective as they scaled up the corpus size using MS MARCO passages. https://t.co/jlQ3N4Pm4J https://t.co/fWnLXHDvkH
— Aran Komatsuzaki (@arankomatsuzaki) May 22, 2023
Custom GPT-4 chatbots are the NEW kid on the block. Here’s how you can have your own with zero coding in two minutes… Introducing Dante: the AI chatbot that anyone can build in 5 minutes: https://t.co/3RyOCvcILX
— Jas Singh (@TheJasSingh) May 21, 2023
Air battle scenes in WWII over the Sea of Japan directed by Steven Spielberg I love the action and war scenes, I hope you like them 💬Prompt in ALT #AIArtistCommunity #aiartcommunity #AIArtwork #AIArt #midjourney https://t.co/gZ5gvDHO0D https://t.co/Q9QDcoMmbl
— MayorkingAI (@MayorKingAI) May 21, 2023
Feynman Lectures on Physics (1964) https://t.co/tgJTe4EgjQ (https://t.co/NMcbx2msot)
— Hacker News 200 (@newsyc200) May 21, 2023
Deploy a PrivateGPT chat-with-documents app in less than 15 mins. PrivateGPT's focus on no-data leaks makes it ideal for enterprises and organizations. Built on @LangChainAI, this app utilizes @huggingface embeddings and GPT4ALL for RetrievalQA. 🧵↴ https://t.co/lV9scOBJu5
— Misbah Syed (@MisbahSy) May 21, 2023
🚨 Latest Latent Space is live! MPT-7B and The Beginning of Context=Infinity with @jefrankle and @abhi_venigalla of @MosaicML https://t.co/KKpo1Dv0JD available at all major podcast retailers (and Hacker News 👀) Personal Highlights: ➜ Discussion of whether @_jasonwei's… https://t.co/0Lh73mKR93 https://t.co/khcLxj7ID8
— swyx.ai (@swyx) May 20, 2023
Part 1 of @LangChainAI course is out now ⚡️ We will understand how apps making million dollars like Jasper, CopyAI work We will understand basics of Langchains, Templates and Chains in details More details below ↓ https://t.co/60HcOR35nd
— Anil Chandra Naidu Matcha (@matchaman11) May 20, 2023
Today was released a ChatGPT PLUG-IN was to generate Midjourney prompts! Here's a short tutorial on how to use it 🧵👇 https://t.co/uUzXGGC7q6 https://t.co/lnqDeh8yxN
— Kris Kashtanova (@icreatelife) May 20, 2023
Mother of all LLM benchmarks! - Use GPT-4 if you need best quality - Use claude-instant-v1 for everything else - Google PaLM2 is nowhere near OpenAI/Anthropic - OpenAI models are painfully slow compared to competition - Open-source models next source: https://t.co/baj7T6LiYl https://t.co/d8KnvEzMI7
— Vlad (@vladquant) May 19, 2023
Excited to announce that the lectures for CS 25: Transformers United V2 (https://t.co/ceK1Hg42EL) are now publicly available 🤩! Our intro lecture featuring @karpathy is out and available to watch here 👉: https://t.co/dWm99wXHCV Spread the word 🤗! #AI #Transformers
— Div Garg (@DivGarg9) May 19, 2023
How to add your face into Midjourney images ❤️🔥 https://t.co/xkhJDniXLB
— Kim (@imkimalban) May 19, 2023
This new research is incredible. Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold. Paper: https://t.co/vsEErv3p2O Project page: https://t.co/PmRe5TfjMd https://t.co/Hh2Yi1zNZU
— Lior⚡ (@AlphaSignalAI) May 19, 2023
Is this the first example of Generative UI? Letting the LLM self-generate the interface is a brilliant idea. https://t.co/pguE5x0Tkt
— Daniel Gross (@danielgross) May 19, 2023
🧵 Ever wanted to talk with your LLM🤖 on some custom data that it wasn't originally trained on? @LangChainAI 🦜🔗+ @pinecone 🌲vectorstore will do all the heavy lifting for you. Here's a simplified explanation using a series of 8 illustrations I made. #GenAI https://t.co/66hjkloFMz
— Eden Marco (@EdenEmarco177) May 19, 2023
Another mindblowing demo. Photo manipulation at an unbelievably Godly level. If you drag the part you want to change, GAN will manipulate it instantly. https://t.co/VvyDyQi7Og
— Barsee 🐶 (@heyBarsee) May 19, 2023
🧪 Favourite's Selection from #midjourney It's Throwback Thursday - I adore countless images, but here is a selection close to my heart. Show me what your favourite images are so far!!! #Prompt 👉🏼 ALT #AIartwork #AIphotography https://t.co/DIVsZea5iN https://t.co/qfsTAEPmuC
— Nicolas Neubert (@iamneubert) May 18, 2023
TUTORIAL: How to make an AI-assisted comic book using Midjourney, ChatGPT and Comic Life 3 https://t.co/eEPQYBrd0s https://t.co/lNS6jIcSfO
— Kris Kashtanova (@icreatelife) May 17, 2023
“The edge will eat the cloud.” Reading Fortinet’s last analyst day transcript/presentation and I liked this formulation. Might take a while though! Also they are the Apple of security. Integrated HW/SW, preference for internal dev, consistently shorted, stock only goes up ✔️
— Gavin Baker (@GavinSBaker) May 17, 2023
Webinar on hallucinations in document question answering tasks starting now Hopefully none of the panelists (@nickscamara_ @awadallah @sfgunslinger @deepset_ai ) hallucinate any answers themselves! https://t.co/hQnEkWFWZs
— Harrison Chase (@hwchase17) May 17, 2023
Clever @google/@stanford paper on LLMs from my brother @jerryweiAI. Performance boosts are great, but there is a more profound insight in this paper that was not explicitly stated: LLMs are trained on human language, but due to the nature of how language was developed (first… https://t.co/KQ4JeGeZ9T
— Jason Wei (@_jasonwei) May 16, 2023
Don't use ChatGPT for academic research. Instead use Bard, the free alternative from Google. Here is how to use Bard smartly for academic purposes:
— Paul.ai (@itsPaulAi) May 16, 2023
CodeT5+: Open Code Large Language Models for Code Understanding and Generation observe sota model performance on various code-related tasks, such as code generation and completion, math programming, and text-to-code retrieval tasks. Particularly, instruction-tuned CodeT5+ 16B… https://t.co/iMbVg40rmu https://t.co/g0tyPL01Hs
— AK (@_akhaliq) May 16, 2023
NEWS: The US Secret Service held an AMA on Reddit and praised blockchain tech https://t.co/LEt5UaW2iR
— Blockworks (@Blockworks_) May 15, 2023
Claude-long (clong to mates) now available in Humanloop. Honestly amazed at how quickly we've arrived at enormous context windows. This feels like a huge research breakthrough. https://t.co/ff3wy7mnuV
— Jordan Burgess (@jordnb) May 15, 2023
We are thrilled to announce an initial integration of @LangChainAI with Docugami. Coupling LangChain with @Docugami’s unique ability to generate a Document XML Knowledge Graph Representation of long-form Business Documents opens the door for... (1/2) https://t.co/ULO0rL514O
— Docugami (@docugami) May 15, 2023
INTRODUCING: How to deploy @LangChainAI to @vercel in 3 steps📚🌩️ 1) 💡Install langcorn 2) 🎨 Create vercel.json 3) Push it to a private GH repo. Demo, Code explained, and GitHub Link ↓ https://t.co/qpGD4Dt524
— Rickyticky Bobbywobbin (@arekusandr_) May 15, 2023
Demystifying Bayesian Models: Unveiling Explanability through SHAP Values by Shuyang Xiang https://t.co/roVFdVPgrA
— Towards Data Science (@TDataScience) May 15, 2023
Models may look stunning on the runway, but have you seen them in space? The combination of their beauty and the vast expanse of the universe is a sight to behold. Prompt in ALT #aiartcommunity #aiart #midjourneyartwork #midjourneyv51 #aigirls https://t.co/1jVfwZ7LoE
— Pixel Cognition AI (@Pixel_Cognition) May 14, 2023
- signal flow diagrams as toy nets for interpretability - diagrams use sequence and monoidal product. add alternate — seq+alt = product+sum = composition+superposition - add kleene star for stuff like deep equilibrium — 1 layer used like iterated fn sys https://t.co/U0e9ILwSam
— maddie 🌸🦜 (@0xmaddie_) May 13, 2023
Humans overestimate the speed at which technological change will happen in the short term and underestimate the impact of that change over the long term. https://t.co/qGROxRpxfs https://t.co/ogd9jFOLFm
— Trevor Loy (@trevorloy) May 13, 2023
Prompt injection explained, with video, slides, and a transcript https://t.co/AszIqOiTyx (https://t.co/uVZLXwYLuv)
— Hacker News 200 (@newsyc200) May 13, 2023
The new 𝚝𝚛𝚊𝚗𝚜𝚏𝚘𝚛𝚖𝚎𝚛𝚜.𝚝𝚘𝚘𝚕𝚜 library from @huggingface is insane! E.g. you can summarize and chat with a PDF in just 6 lines of code, including import statements! 🤯 The best part? It's all open-source, from start to finish! 🤗 https://t.co/W1L1J4UzRT
— DataChazGPT 🤯 (not a bot) (@DataChaz) May 13, 2023
Introduction to Computer Vision is a free introductory course @cornell_tech by @Jimantha. The course covers topics from image formation to 3D shape reconstruction, object/face detection to deep learning. Learn here: https://t.co/DEOl5dZlTV https://t.co/DX35lxMdwt
— TuringPost (@TheTuringPost) May 12, 2023
By now you may have seen some hubbub about @MosaicML’s MPT-7B series of models: MPT-7B base, MPT-7B-Instruct, MPT-7B-Chat, and MPT-7B-StoryWriter-65k+. These models were pretrained on the same 1T token data mix. In this 🧵I break down the decisions behind our pretraining data mix https://t.co/ifoWjqUU4S
— Matthew Leavitt (@leavittron) May 12, 2023
Comparing @Google Bard, @AnthropicAI Claude, and @OpenAI GPT-4 performance on tricky prompts:
— Matt Shumer (@mattshumer_) May 12, 2023
Google releases Bard to the world – leaves EU behind @bimedotcom @evaXvar @Khulood_Almani @danfiehn @tobiaskintzel @chidambara09 @sonu_monika @theomitsa @BetaMoroney @Analytics_659 @Shi4Tech @FmFrancoise @enricomolinari @enilev @sallyeaves @IanLJones98 https://t.co/zzQgJbTC32
— Knut Jägersberg (@JagersbergKnut) May 12, 2023
Re-made a chatbot app using @LangChainAI @pinecone @trychroma @streamlit @OpenAI. However, I wasn't thrilled about writing the https://t.co/M4OncT3rej file. So, I let the app ingest the app file and asked it to write the README file for me. And voila, it worked like a charm! https://t.co/GxuP0rbW4E
— ZenNE (@ZenNEchan) May 11, 2023
ChatGPT has now a big problem. Google just updated its free competitor, Bard. Here are 8 things impossible on ChatGPT but that Bard can do (for free):
— Paul.ai (@itsPaulAi) May 11, 2023
8 billion lives will change because of AI. But those who think large models are the end-game don't know what's happening. The real game-changer coming after every industry is the applications we are building right now. https://t.co/J49iwMKcg7
— Santiago (@svpino) May 11, 2023
Googles Keynote 🤯 for AI. Here's a brief overview: About 25 different AI products. Biggest LLM news Palm 2 - 4 different sizes, gecko (runs on mobile) , otter, bison, unicorn Palm med - tuned for Medical purposes. Can analyze x-rays to help radiologists. - Palm Security
— Sully (@SullyOmarr) May 10, 2023
I rarely pay attention to macro forecasts. I’d rather spend time looking for great businesses selling cheap. However, one person I listen to on macro is Stan Druckenmiller. Just finished listening to his presentation at the USC 🧵👇 1/15 https://t.co/YuxWietEKh
— Hidden Value Gems (@HiddenValueGems) May 10, 2023
🧪 House of Prompts Vol. 1.5 Last week, we shared the first part of the House of Prompts series. Shortly after, #midjourney 5.1 was released, and we wanted to update the post using the new model. Check out the rooms below 👇🏼 #aiphotography #architecture #interiordesign https://t.co/8YBrLsck3k https://t.co/reL0aMetg2
— Nicolas Neubert (@iamneubert) May 9, 2023
AI video has started to produce mindblowing results and could eventually disrupt Hollywood. (PT7) Here are the best AI videos I've found:
— Nathan Lands (@NathanLands) May 9, 2023
In this article you will see the six components used to build the Agent. @LangChainAI @Langflowai #LangChain #LangFlow https://t.co/ifE1amb20G
— Cobus Greyling (@CobusGreylingZA) May 8, 2023
I really think everyone should read the GPT-4 whitepaper released by OpenAI. At a minimum: Pages 4-8: System capabilities Page 48: Chart showing you why guardrails are needed Pages 54-56: Emergent behavior Pages 58-59: Economic impact and acceleration https://t.co/XXlQvvckgD https://t.co/dGcVc0lDsr
— Ethan Mollick (@emollick) May 8, 2023
A list of open source LLMs available for commercial use by @eugeneyan This github repo contains ⦿ Open source LLMs for commercial use along with their licence ⦿ List of open source LLMs for coding ⦿ Evals and LLM datasets for fine-tuning to be added https://t.co/BdjuVFianT https://t.co/9p57fUf1Ml
— SRK (@sudalairajkumar) May 8, 2023
https://t.co/1x9bYKKbb8
— Gopi Vikranth (@GopiVikranth) May 8, 2023
Open source LLMs are all the rage now. @nuwandavek & I finetuned a RedPajama 3B OSS LLM using LoRA on a small dataset of 200000 dad jokes from Reddit. We also built a small flask app that renders it in Joe Biden’s voice. Play with it here(for next 24h) https://t.co/T8kT4kJCf6
— Keerthana Gopalakrishnan (@keerthanpg) May 7, 2023
“our 770M T5 model outperforms the 540B PaLM model using only 80% of available data on a benchmark task.” https://t.co/gInbAM8VTd
— Nathan Benaich (@nathanbenaich) May 7, 2023
Not sure how to create the latest Louis Vuitton puffer jacket drop? /imagine: large format film photo of a european model wearing a Louis Vuitton branded puffer jacket --ar 4:5 It’s that simple in V5.1 … don’t let anyone tell you otherwise. Run your own variations and… https://t.co/sRaLyR7sJL https://t.co/hWZ1OHEIeF
— JOSHUA TEPLITZ (@josh_tep) May 7, 2023
Indexes in @LangChainAI are confusing. But I've broken it down into bite-size chunks. • Basic Information • Document Loaders • Retrievers • Vector Stores • Text Splitters Save this ↓ https://t.co/UPQpPKQjhZ
— peter! 🥷 (@pwang_szn) May 7, 2023
Here's the ChatGPT Ultimate Prompting Guide Important elements to include: 1. Tone 2. Format 3. Objective 4. Context 5. Examples (Source: Reddit) https://t.co/62zoHphuZf
— Aadit Sheth | Mr. Prompts (@aaditsh) May 6, 2023
With the heat wave, figured might as well go somewhere more enjoyable with sun, while keeping a slightly different royal theme for the weekend. Enjoy Monaco. #synthography #AIart #midjourney #midjourney51 #aiphotography #postphotagraphy #aiartcommunity https://t.co/xQ0ZGAlbeh
— Helena Manzanita (@HelenaManzanita) May 6, 2023
Evaluation of LLM question+answering chains can be challenging: here's @huggingface space to automate this. Upload doc(s) and select a QA chain configuration you want to test. The app builds the chain (w/ @LangChainAI), grades it, and logs results for you. https://t.co/utr7S42f88 https://t.co/puVFGdSQ9H
— Khaled Hammadi (@KhaledHammadi32) May 6, 2023
Oh I forget to mention. You can train your own local AI model using this framework. I will show you how for FREE in the https://t.co/Ruqey26qLY article I am writing and testing now. Your life will never be the same. Join us, we are the pioneers. https://t.co/HiYsbJxQZ5 https://t.co/jHILpnk3Vq
— Brian Roemmele (@BrianRoemmele) May 6, 2023
This was quite a week in AI! Google's "no moat" leak and amazing new releases by @togethercompute @MosaicML @BigCodeProject our blog talks about how AI's technical moat is shrinking and why it's good to be optimistic about open source: https://t.co/SeI0FEQBpM https://t.co/FimAXzsNym
— hazyresearch (@HazyResearch) May 6, 2023
AutoML-GPT, which employs GPT as a bridge to diverse AI models and dynamically trains models with optimized hyperparameters. https://t.co/KXJycYgUC1 https://t.co/tx0NaMrOz7
— Carl Carrie (@🏠) (@carlcarrie) May 5, 2023
The first RedPajama models are here! The 3B and 7B models are now available under Apache 2.0 license, including instruction-tuned and chat versions! This project demonstrates the power of the open-source AI community with many contributors ... 🧵 https://t.co/msO4afBQEK https://t.co/ekLeidmd0q
— Together (@togethercompute) May 5, 2023
Get it while it's hot! The @MosaicML MPT-7B -Instruct model is trained on the databricks-dolly-15k instruction tuning corpus. LLaMA-grade language capabilities with Dolly-like instruction following. Ship it! https://t.co/i0jXBdjyV6
— Mike Conover (@vagabondjack) May 5, 2023
https://t.co/iNdAwUv6lt
— Patrick Collison (@patrickc) May 5, 2023
My thoughts on the now famous Google leak doc: https://t.co/hK2KcMTEDB 1. Open source AI is winning. I agree, and that is great for the world and for a competitive ecosystem. In LLMs we're not there, but we just got OpenClip to beat openAI Clip and Stable diffusion is better than… https://t.co/mU6CismgLN
— Alex Dimakis (@AlexGDimakis) May 5, 2023
Stunning @BigCodeProject! A trillion tokens in 80 languages, huge context window, and step function improvement in quality. 38% improvement on HumanEval Pass@1 vs Replit's replit-code-v1-3b. Nearly doubles the performance of models in the Salesforce Codegen family. https://t.co/atUzP5lT3I
— Mike Conover (@vagabondjack) May 4, 2023
Text-to-websites tool is the next disruption. Dream enables anyone to build & deploy functional web applications with prompts instead of code. Just imagining how good these are going to get in a couple of months 👇 https://t.co/N0ggXdHppT
— Barsee 🐶 (@heyBarsee) May 4, 2023
Open LLMs are pushing harder than ever, once there is an equivalent to LLaMA with commercial usage license I do expect a lot of stiff competition from privately fine tuned versions https://t.co/N0KlQbWBaF
— anton (@abacaj) May 4, 2023
LLMs/AlpacaGPT4-LoRA-7B-OpenLLaMA Soon there will be high performance LLMs with permissive licenses. It won't take long until the fully trained openllama get's a fine tune on OpenAssistant data. Later we will have 30b redpajama on that. https://t.co/9IfXVMZyuI
— Knut Jägersberg (@JagersbergKnut) May 4, 2023
Two Snoop Dogg snippets from Milken today. I was sitting up front, and the whole thing was bananas. First, Snoop Dogg on streaming media pricing: "If you can get a billion streams, why can't you get a million dollars? Who the f**k run the streaming industry? Are you in here?" https://t.co/e6czGzDS84
— Paul Kedrosky (@pkedrosky) May 4, 2023
FREE A.I. COURSES Introduction to A.I. https://t.co/AjyCb2wYI6 Learn prompting https://t.co/Klnt3kqUqo Promoting with ChatGPT (using API): https://t.co/6Wu9OvQr6x Python / Machine Learning: https://t.co/8GTaibKTP8
— Kris Kashtanova (@icreatelife) May 4, 2023
Agents are great! But for true ongoing learning, the agents need to actually learn, ie update the model, not just remember. Been thinking on it for a bit, so here's Autotune_GPT, with recursive self-improvement. Every 10 prompts it updates its model! https://t.co/MKLXP3Jglg
— rohit (@krishnanrohit) May 3, 2023
Distill or Annotate? How to Most Efficiently Use a Fixed LLM Budget -Extensive experiments on 6 diverse NLP tasks -Distilling LLM (11B) to small language model (60M) is more cost-effective -Compared to annotating more data to directly train small model https://t.co/9lJ09OrcAU https://t.co/c38b9z41IR
— John Nay (@johnjnay) May 3, 2023
📢 Meet CSV Agent 🙌 It's specifically optimized to interact with tabular data and can perform more complex calculations through LLM 🤖🚀 It's now LIVE on @finetuner_ai - check it out 👀🔥🔥 #NoCode #AutoGPT #AI #buildinpublic @LangChainAI #GPT4 #MachineLearning https://t.co/zPNWLQOQDJ
— 0xAlbert_S3 (@0xAlbert_S3) May 3, 2023
#AgentGPT recently hit 20k stars on GitHub 🚀 Its been an incredible ride watching this project grow. A big shoutout to everyone that's helped contribute the past few weeks https://t.co/bLu2PaKMs0
— Asim Shrestha (@asimdotshrestha) May 2, 2023
Who had this on their 2023 bingo card - infighting among activists? https://t.co/QNwgEdwoCO
— Scott Kupor (@skupor) May 2, 2023
Excellent TED talk from Sal Khan: - many inspiring examples of GPTs finetuned into socratic tutors, assisting without giving away answers. - none of it "out of the box", requires prompt engineering, finetuning, data collection, iteration. - sense of barely scratching the surface. https://t.co/LITvcRk0Jw
— Andrej Karpathy (@karpathy) May 2, 2023
What is prompt injection? How does it work? How dangerous is it? What can be done to prevent it? Tune into our webinar tomorrow to find out! We've got some truly awesome guests to help shed some light https://t.co/w2PafcT8A1
— langchain (@LangChainAI) May 1, 2023
We're announcing improvements to our callbacks system, which powers logging, tracing, streaming output, and some awesome third-party integrations Blog: https://t.co/pUgGF4NT12 Python docs: https://t.co/KEmiRQaLd1 JS docs: https://t.co/8cMApHyLiL 🧵
— langchain (@LangChainAI) May 1, 2023
Charlie Munger’s Investment Principle Checklist should be known by any Investor. These two pages summarize all Investment Philosophy you need. Today we will walk through it (+PDF at the end)👇🏼 https://t.co/Bin5a4UamH
— Daniel (@MnkeDaniel) May 1, 2023
