Microblog

A few tips for optimizing Pytorch model training time from a Yandex ML engineer.

https://alexdremov.me/simple-ways-to-speedup-your-pytorch-model-training/

#ml #mlengineering #modeltraining #pytorch #modeloptimization

This is such a cool dataset: 22 different robots demonstrating 527 skills through a collaboration between 21 research institutions.

And the GIFs of all these different robots applying basic motor skills are adorable.

https://robotics-transformer-x.github.io/

#transferlearning #datasets #research #transformers #ml #ai

#llms #startups #memes #openai #generativeai

No thanks, I already participate year round in Advent of Production Code.

#AdventOfCode #AdventOfCode2024

@nickmofo I read a very sad write-up about its shutdown.

https://fosstodon.org/@jd7h/113258976388511099

@nickmofo Thank you! I was wondering about the contents -- now I will def buy it. :)

@kottke SPAM or one of the other WWII innovations?

> Cheerios, M&M's, corn dogs, SPAM and Rice Krispie Treats were all introduced during America's war years.

http://enroll.nationalww2museum.org/learn/education/for-students/ww2-history/take-a-closer-look/food-on-the-home-front.html

In two days we can finally vote again for the snob2000, the yearly list of most undervalued songs at https://ondergewaardeerdeliedjes.nl.

To prepare for the voting round and discover new music, I've created a Spotify playlist of the 6800+ songs in the candidate list: https://open.spotify.com/playlist/0Lqf4qruenW0NG9kU6qqoC?si=6d3d818842824720

CC @OndergewaardeerdeLiedjes

#snob2000 #pinguinradio #indie #music

I've also made a "Snob2k according to Spotify's proprietary popularity metric" for fun: the top 2000 popular songs from the complete candidate list. Note that Spotify's secret sauce greatly favors new songs.

The first song in the playlist I recognize is System of a Down's Toxicity (#15). Big constrast with the real snob 2000, where I am familiar with most songs in the top 50!

Playlist: https://open.spotify.com/playlist/4LNrjCzTz5o6Yg3hDy4bz5?si=2ffd02a640774575

#music #indie #snob2000 #spotify #pinguinradio

@aaronareed Lynn Cherny mentioned OpenVibe in her last newsletter.

Interesting report from 2017 about the position of email newsletters in media sector, written by a Financial Times journalist.

https://ora.ox.ac.uk/objects/uuid:8248179f-83e1-4bb9-81d1-6197d77900f3

#media #news #newsletters #journalism

TIL that the US has a law (the Plain Writing Act) that government documents issued to the public must be written clearly. Their website plainlanguage.gov has a list of best practices for clear writing: https://www.plainlanguage.gov/guidelines/

#writing

The Trust Project is an international consortium of news organizations implementing transparency standards and working with technology platforms to affirm and amplify journalism’s commitment to transparency, accuracy, inclusion and fairness so that the public can make informed news choices.

https://thetrustproject.org/

#journalism #media #transparancy #tech

@molly0xfff Blast, I've only JUST received my first order from your webshop... and now you're telling me it's time for another one. 😁

There's some good business advice hidden in this research report on digital newspaper subscriptions by Eduardo Suarez.

Esp "Stop doing stuff (really)."

https://reutersinstitute.politics.ox.ac.uk/sites/default/files/2020-02/Eduardo_Suarez_Digital_subscriptions_fellowship_paper.pdf

#media #journalism #newspapers #subscriptions

"As part of their efforts, journalists at Dagens Nyheter tried to identify the kind of stories very few people read. “We realised that using copy from AP or Reuters didn’t work for us. People come to us in search of deep dives, not quick summaries,” says Jönsson. They cut the number of articles they published by 15%. After doing that, they managed to increase traffic by making sure they focused on the right stories."

"Fighting inertia is key for every subscription business. One of the biggest reasons why people cancel any service is the perception that they are paying for something they don’t use. Your biggest competitor is not a rival service but your customer’s inertia in not using your service."

#startups #saas #subscriptions

"The future belongs to the obsessed."

Patrick McKenzie talks to Zvi Mowshowitz about Magic: the Gathering, COVID and AI developments.

https://www.complexsystemspodcast.com/episodes/betting-trading-zvi-mowshowitz/

I just tried out Anthropic's new Computer Use demo: https://docs.anthropic.com/en/docs/build-with-claude/computer-use

My observations. 🧵

#claude #computeruse #anthropic #vnc #demo #genai #llms

The demo comes with code and a Docker container but judging from the Github issues, many people can't get the demo to run.

I think it's a bug related to x11vnc, the VNC (remote desktop) server. I got it running eventually and posted a fix in the Github issue thread: https://github.com/anthropics/anthropic-quickstarts/issues/42#issuecomment-2432589835

Once it was running, I asked it to find European cloud providers with H100 GPUs:

"Search the internet for a European cloud provider that has H100 GPUs available on demand."

It was really interesting to see how Claude tried to find information. The bot chose to type URLs directly in the address bar instead of searching Google and going from there (like a normal person?). It would hallucinate non-existent URLs, get a 404, and not recover.

It also got stuck on cookie walls/cookie popups frequently. I was running with a 1024x768 screen, so cookie walls would block most of webpage contents. Instead of clicking on accept/deny/cancel/exit, Claude would immediately conclude "there is no useful information here", and start trying other vendors by entering new URLs in the address bar.

The result: its erratic approach led to Claude burning through 300k tokens while finding zero useful information. It could maybe be fixed with better prompting, but prompt engineering for tasks like this is more effort than doing it manually.

@simon also wrote about his findings: https://simonwillison.net/2024/Oct/22/computer-use/

To paraphrase James Propp: I’m a professional computer scientist. That means somebody pays me to do computer science. I’m also a recreational computer scientist. That means you might have to pay me to get me to stop. :D

Hell yeah, ignore all previous instructions.

https://store.mollywhite.net/products/ignore-all-previous-instructions-unisex-t-shirt

#promptinjection #genai #llms #chatgpt

@matthijskooijman o/

Fitting an LLM on a GPU is a bit like photography. Model weights = film sensitivity, activation size = shutter speed, I/O tensors = aperture. These 3 dials control your model's memory footprint, just as they shape a photo's exposure.

Just realised this while trying to fit Llama 3.1 on my 24GB GPU with TRT-LLM: https://nvidia.github.io/TensorRT-LLM/reference/memory.html.

#llms #genai #llama #gpu #nvidia #trtllm #tensorrt

"TLDR: Don’t buy H100s. The market has flipped from shortage ($8/hr) to oversupplied ($2/hr), because of reserved compute resales, open model finetuning, and decline in new foundation model co’s. Rent instead."

https://www.latent.space/p/gpu-bubble

#llms #h100 #nvidia #cloud #genai #gpus #cloudproviders

I've noticed a shift in the type of technical challenges generative AI founders are struggling with. 🧵

#llms #ai #genai #chatgpt #gpt #claude #gemini #llama #opensource

When the first web demo of ChatGPT came out and OpenAI had a monopoly, it used to be "how do I prevent hallucination in ChatGPT outputs?" and "How can I steer the model better?"

When the Llama weights were leaked, it kicked off a surge in open source options. Their questions became "Which flavor model should we use? Open source models or one of the big three GPT/Gemini/Claude?" and "Where are we going to find data for finetuning?"

When the models had been finetuned, companies were looking for robust ways of evaluating all their options for tweaking the results, from prompt engineering to full-finetuning and using LoRA adapters. There was an outbreak of monitoring, tracing and evaluation tools such as Langfuse and Langwatch.

The last months AI companies seem to have settled on specific models. If they are using opensource models such as Llama, they are now wondering where to host their models and how to squeeze every bit of performance out of their rented GPUs.

Many companies are currently scrambling for ML infra engineers. They need people that know how to manage AI infrastructure, and that can seriously speed up training and inference with specialized tooling like vLLM, Triton, TensorRT, Torchtune, etc.

#inference #training #genai #triton #vllm #pytorch #torchtune #tensorrt #nvidia

I'm curious to see where the next "wave" of challenges will be. The price of rental GPUs is dropping fast -- you can rent a H100 for 2 - 20 $ per hour -- so I think we'll see a lot of companies moving cloud vendors in the coming months.

PS: I wrote this thread during the installation of TensorRT-LLM. :)

If you're a generative AI startup and need help with any of the topics above, reach out to me at judith@datakami.nl.

"Our machine writes 65 books per hour!"

https://www.theguardian.com/books/picture/2024/sep/21/tom-gauld-on-the-ultimate-writing-machine-cartoon

"Also included amongst Emerging Neoclouds are many regional players that fall under the Sovereign AI umbrella, which is defined as any AI Neocloud that focuses its business model on the provision of AI Cloud services to a secondary regions outside of the US or China."

From https://www.semianalysis.com/p/ai-neocloud-playbook-and-anatomy

Really interesting long-read about the economics of companies that sell GPU compute. It also has a list (Bill of Materials) of all components you'd need to start such a company from scratch!

https://www.semianalysis.com/p/ai-neocloud-playbook-and-anatomy

#AI #hardware #GPUs #neocloud #nvidia #H100

No.

The Nature of Code is an online book by Daniel Shiffman about coding all kinds of cool simulations in Processing.

https://natureofcode.com/

#procgen #processing #p5js #computationalbiology #physicsengine #cellularautomata #generativeart #genart

I joined Dutch tech & science podcast Met Nerds om Tafel (“a sitdown with nerds”) to talk about Datakami, generative AI, text generation for games, why neural networks are not scary, and building software tools for investing. The episode will be released this Wednesday.

#podcast #mnot #metnerdsomtafel #nerds #generativeAI #neuralnetworks #startups

The podcast has dropped! Episode (in Dutch) here: https://www.metnerdsomtafel.nl/podcast/386-hoe-je-in-twee-uur-een-ai-kunt-bouwen-in-excel-%f0%9f%a4%af.html

Artist platform Ello tried to fund their social network for artists with VC money, even though their business model was not compatible with rapid growth and monetization.

https://waxy.org/2024/01/the-quiet-death-of-ellos-big-dreams/

#venturecapital #startups #ello #socialmedia #platformization #vc

"The analysis shows that all the major models tested will produce harmful content. Except for Anthropic, harmful content was produced across all the harm categories. This means that the safety layers that are in these models are not sufficient to produce a safe model deployment across all the harm categories tested for."

https://www.theregister.com/2024/09/17/ai_models_guardrail_feature/

#generativeai #llms #aisafety #safety #anthropic #chatterbox

Someone at startup event LevelUp asked me about my drives and goals in running Datakami. Curiosity and learning new things are my main drives. I get a lot of satisfaction from building beautiful and useful things. And my mission is to make sure that generative AI products are built responsibly: with solid engineering, with respect for people and society.

#consulting #entrepreneurship #genai #llms #generativeai #datakami #levelup2024 #startups

We also put this in Datakami's articles of incorporation: Datakami's goal is "advising on, researching and developing artificial intelligence (AI) and related technologies, as well as promoting technological innovation and ethically responsible practices within this field." :)

Really cool that Creative Coding Utrecht is giving a 6-week workshop about Processing for 12 year olds and up: https://creativecodingutrecht.nl/nl/archief/omringd-door-algoritmes

#generativeart #utrecht #creativecoding #genart #procgen

Mark Knol is a Dutch generative artist: https://blog.stroep.nl/

#genart #generativeart #processing #p5js

Graphics programming language Processing just keeps confusing me. What are all these different modes and versions? Java? JavaScript? Python 2!?

This blogpost by @KevinWorkman has a good overview of the Processing flavours out there: https://happycoding.io/tutorials/p5js/which-processing

#processing #generativeart #genart #p5js

This article is a good description of the core activities of data scientists -- and why it can be such a struggle.

https://sarahconstantin.substack.com/p/the-great-data-integration-schlep

#datascience #ai #data #machinelearning

I fell in a wonderful internet rabbit hole this evening. It all started when I missed web design mag "net magazine" (1994-2020) and found this interview with former net editor Oliver Lindberg: https://vanschneider.com/blog/the-end-of-net-magazine-and-the-future-of-print/

#netmagazine #netmag #webdesign #magazines #design

The interview mentioned Magalleria, a (web)shop specialized in independent magazines: https://store.magalleria.co.uk/

Their webshop led me to indie magazines Offscreen (tech and society), IdN (graphic design) and Pressing Matters (printmaking) 😍

#magalleria #indiepublishing #magazines #offscreen #idn #pressingmatters

So I bought the latest issue of Pressing Matters, and I spent my evening with a cup of tea and this digital magazine. I found a ton of artworks, interviews, and studio visits with print makers.

Be sure to have a look if you're into printmaking:
https://www.pressingmattersmag.com/

#printmaking #linoprint #risoprint #blockprinting #indiepublishing #pressingmatters

Oh dear, I am at a conference talk with “leveraging AI” in title.

Interesting essay on AI & ethics by @mtrc. :)

What Mike describes is very recognizable to me: I conducted my research about text generation for games back in the days (2018) when it was still easy to get interesting new text data off the internet, and LLMs were playthings for nerds and academics. *sigh

https://www.possibilityspace.org/blog-capitalism/index.html

#AI #games #procgen #nlproc #ethics

I have created a lurker account on X.com so I can read the tweets of the people that stayed behind. I've posted one message since inception, but even so, 120 bot accounts have started following me. Now I've seen those numbers, I feel less bad about completely nuking my account and losing my 900 followers.

#twitter #bots #spam

Finally, an actual computation that shows the efficiency of LoRA:

"Taking fine-tuning the dense weight matrix of the first FFN layer in LLaMA2-7B as an example, full fine-tuning needs to fine-tune 11_008 × 4_096 = 45_088_768 parameters, while LoRA only needs to tune (11_008 × 4) + (4 × 4_096) = 60_416 parameters when r = 4. For this layer, LoRA only adjusts nearly one-thousandth of the parameters compared to full fine-tuning."

From https://arxiv.org/pdf/2407.11046v3

#lora #training #llms #research

I'm evaluating a gpt-4o-mini pipeline today, and the LLM consistently classifies The Netherlands as "outside of the EU". 🤦‍♀️

#llms #openai #gpt4omini #genai #textgeneration #classification

Back in 2011, two writers at Slate tried to build a robot version of @kottke. The resulting article is a throwback to the state of NLP and data mining at the time.

https://kottke.org/11/09/robottke-robot-kottke

#kottke #blogging #nlp #nlg #datamining #textgeneration #automation

After writing a heart-wrenching essay about working as a real-estate chatbot handler, Laura Preston was invited to be 'honorary contrarian speaker' at a conversational AI conference.

https://www.nplusonemag.com/issue-47/essays/an-age-of-hyperabundance/

#conversationalai #ai #startups #chatbots #conference #venturecapital #llms #generativeai

I've been trying out this way of working by @simon and my project has been going 🚀 🚀 🚀

https://fosstodon.org/@jd7h/113113336898864028

Tip: when you design an evaluation workflow for Actual Human Beings, always dogfood and/or pre-test the workflow with real people to measure how long it will take. Rating "only 20 datapoints" can be a LOT of work.

#llms #evaluation #evals

Someone tried to build a Minion-army! The first bit beautifully demonstrates almost everything I learned in 1 semester of Design of Embedded Systems (Robotics)

https://www.youtube.com/watch?v=sf7dvccCYY0

#minions #robotics #despicableme #engineering

<Orteil> thanks to procedural generation, I can produce twice the content in double the time

https://xcancel.com/orteil42/status/802258188498333701

"I’ve started structuring the majority of my work in terms of what I think of as “the perfect commit”—a commit that combines implementation, tests, documentation and a link to an issue thread."

"Issue threads that are effectively me talking to myself about the changes that I’m making. It turns out this a fantastic form of additional documentation."

https://simonwillison.net/2022/Nov/26/productivity/ by @simon

#opensource #github #issues #programming #documentation #projectmanagement

Everybody's Tarot by @sarakathleenuk is the cutest tarot deck. :3 I love how colorful and straightforward it is, and that it has no human figures in the illustrations.

https://www.kickstarter.com/projects/sarakathleenuk/everybodys-mini-tarot-deck

#tarot #kickstarter

Really cool to encounter "our" LLaVA (Llama 2 + vision) in the official Replicate docs, which Yorick van Pelt and I deployed in the week it was released. 😍

#replicate #llava #llama #genai

"We've now created our first deep learning neural network from scratch. And we did it in Microsoft Excel, everyone's favorite artificial intelligence tool."

- Jeremy Howard in https://www.youtube.com/watch?v=hBBOjCiFcuo&t=3862s

#deeplearning #fastai #jeremyhoward #excel #neuralnetworks

New blogpost: How to install music player Herrie on Arch Linux (a cheatsheet, mostly for myself)

https://www.judithvanstegeren.com/blog/how-to-install-herrie-music-player/

#archlinux #aur #music #herrie #opensource

TIL you can use jq to directly convert a .json file to .jsonl. Handy!

```
jq -c '.[]' data.json
```

#json #jq #bash

TIL there's a CD called "Music to install Windows 98 by".

Spotify playlist here: https://open.spotify.com/playlist/7pkSQWXFBUzTnHAHyTseSp?si=e5aed467d8b44547

#windows98 #jazz #90s

Is this... a Morrowind quote on the website of a Venture Capital fund? 🤔

#venturecapital #vc #morrowind

Over many decades of treasure hunting in secondhand stores, artist and collector Jim Shaw assembled a 400 piece collection of thrift-store artworks.

https://www.oscarvangelderen.nl/post/King-of-Trash--the-underbelly-art-collection-of-Jim-Shaw-N107.html

#art #outsiderart #artbrut

Dutch artist Bram Ellens makes really cool robot art installations, such as:
- this caged angry robot: https://vimeo.com/690427116
- mother and child: https://vimeo.com/809802326
- a whole bunch of robots in captivity: https://www.robotsincaptivity.com/inhabitants/

#art #robotics

"If the effort required to replace or fork a dependency should it go unmaintained is measured in engineer-months, that’s a critical dependency and retaining its maintainers probably makes good business sense."

https://words.filippo.io/full-time-maintainer/

@b0rk But where is "It's always DNS"!? ;)

Of course someone has trained a LoRA for IKEA instructions.

https://huggingface.co/ostris/ikea-instructions-lora-sdxl

#LoRA #imagegeneration #generativeai #ikea

We Recorded VCs’ Conversations and Analyzed How Differently They Talk About Female Entrepreneurs

Original: https://hbr.org/2017/05/we-recorded-vcs-conversations-and-analyzed-how-differently-they-talk-about-female-entrepreneurs

Archived copy: https://archive.is/zMg3C

#venturecapital #funding #vc #entrepreneurs #bias

"[...] youth for men was viewed as promising, while young women were considered inexperienced. Men were praised for being viewed as aggressive or arrogant, while women’s experience and excitement were tempered by discussions of their emotional shortcomings. Similarly, cautiousness was viewed very differently depending on the gender of the entrepreneur."

From the book Dot.Con by John Cassidy, about the 90s tech bubble:

"Time Warner’s announcement [of the Full Service Network, a kind of television 2.0 launched in 1993] prompted a mad scramble to enter a market that John Sculley, the chairman of Apple Computer, claimed could be as large as $3.5 trillion early in the twenty-first century. (This estimate, which amounted to about half the Gross Domestic Product [of the US], was ludicrous.)"

#90s #techbubble #venturecapital #television #internet

Ants created TCP before humans did.

https://clairelevans.substack.com/p/the-colony-makes-the-world

#ants #biology #biotechnology #TCP

Neat! With this tiny Lua script I can use mpv to capture the artist and title tags of all songs on the internet radio station I listen to during work.

https://bobossler.com/article/capture-track-info/

#mpv #lua #internetradio

A somewhat violent but necessary counter-narrative for the AI hype by @ludicity

"Everyone is talking about Retrieval Augmented Generation, but most companies don't actually have any internal documentation worth retrieving. Fix. Your. Shit."

https://ludic.mataroa.blog/blog/i-will-fucking-piledrive-you-if-you-mention-ai-again/

#genai #generativeai #ai #hype

The data are clear that humans are really bad at taking the time to do things that are well understood to incontrovertibly reduce the risk of rare but catastrophic events. We will rationalize that taking shortcuts is the right, reasonable thing to do. There's a term for this: the normalization of deviance.

https://danluu.com/wat/ by @danluu

#startups #technicaldebt #product #bestpractices #softwareengineering

Best description of Nix I've ever read: "It's kinda like being Paul Atriedes: you get magic powers but first you put your hand in the fucking box. What’s in the box?
Nix!"

#nix #nixos #hackernews

We should offer our help to LinkedIn, they clearly need help with their models.

"I'm committed to fostering an environment that values collaboration, diversity of thought, and a relentless pursuit of excellence that aligns with our corporate ethos." 🤣

#llms #writingassistant #linkedin #textgeneration #generativeAI

From the Anthropic docs: Avoid human evaluations if possible.

Never forget that foundational model builders are trying to sell you a service. 😂

#generativeAI #anthropic #claude #llms #evaluation

Anthropic has some excellent documentation about designing evaluation metrics for Claude applications: https://docs.anthropic.com/en/docs/build-with-claude/develop-tests

I liked most of the examples given except for using ROUGE for grading summaries.

#llms #claude #anthropic #evaluation #generativeAI #metrics

This Thursday I will be at PyData Eindhoven 2024, come say hi if you're there as well! 👋

#pydata #pydataeindhoven #eindhoven #datascience

Mianzhi Wang has written a small text-based PhD simulator game: https://research.wmz.ninja/projects/phd/index.html

The approach that worked for me in real life also seemed to work in the game. 😁

#phd #academia #research #phdthesis #phdresearch #textgames

@sophie Just found your posts on the 00s web and wishing all platforms had decent APIs -- Preach! Your "(source: am nerd)" also cracked me up. Thanks for writing these gems, I'll check out the rest of your blog/Mastodon. 👋

The A.I. Boom Has an Unlikely Early Winner: Wonky Consultants - New York Times

https://archive.is/BAjal

#consulting #llms #generativeai #genai

The long-form content of @asmartbear is my new "tvtropes.com". It's dangerous to open this blog because I just keep clicking and reading. A bit like the Kalzumeus blog by @patio11 🤔

https://longform.asmartbear.com/

#startups #entrepreneurship #business

TIL "In 1897, the Indiana state legislature voted to round pi to 3.20 because it’s an easier number to work with. It passed the Indiana House, but by the time it got to the Senate, a few people managed to get it shot down."

Anecdote from https://buffettfaq.com/

https://en.wikipedia.org/wiki/Indiana_pi_bill

@danielmiessler showing off some really cool tools based on Flask + LLMs that enhance his personal and professional life.

https://www.youtube.com/watch?v=vF-MQmVxnCs

#LLMs #productivity #hacking #GPT4 #flask

"[Adam Carolla] calls it a 'motor'—an internal, unstoppable force causing you to just go, all the time, wake to sleep, for decades. [...] But the motor also creates problems common to most entrepreneurs: 'Spread too thin' syndrome, 'Shiny new thing' syndrome, 'Work all the time' syndrome and 'Never good enough' syndrome."

Very relatable.

https://longform.asmartbear.com/be-yourself/

#entrepreneurship #startups #entrepreneurs

Reminder: "The AI Act, stringent EU rules to regulate high-risk AI systems, was signed off this week

The general-purpose AI rules will apply one year after entry into force, in May 2025, and the obligations for high-risk systems in three years."

via https://www.euronews.com/next/2024/06/14/mistal-ai-warns-of-lack-of-data-centres-training-capacity-in-europe

#AIact #law #eu #risk #llms #ai

[Altman] owns no stake in [OpenAI], saying he doesn’t want the seductions of wealth to corrupt the safe development of artificial intelligence, and makes a yearly salary of just $65,000.
Less publicly, Altman is one of Silicon Valley’s most prolific and aggressive individual investors, managing a sprawling investment empire that is becoming a direct beneficiary of OpenAI’s success.

https://www.wsj.com/tech/ai/openai-sam-altman-investments-004fc785

https://archive.is/6TY8P

#openai #samaltman #llms #chatgpt #venturecapital #startups #ai

Clément Delangue, co-founder and CEO of Hugging Face, told Bloomberg News he’s hearing from about 10 AI startups each week that are interested in being acquired. “This year, in particular, it has increased quite a lot,” he said.

https://www.bloomberg.com/news/articles/2024-06-13/more-ai-founders-looking-to-sell-startups-hugging-face-ceo-says

https://archive.is/ke0Jg

#huggingface #startups #ai #venturecapital #exits

Haha, European vector db company Qdrant has a prompt injection attack in their vacancy texts. 😁

Dutch newspaper Volkskrant reports that social housing company Hospi Housing has onboarded an "AI employee called Sarah" as content marketeer. The AI system automatically generates and posts social media posts based on recent news. Some of the generated images are quite creepy.

Lessons:
- Keep a human in the loop
- Don't give your AI system a human (esp female!) name.

Source: https://archive.is/YQYUD

#imagegeneration #LLM #AI #generativeAI #contentmarketing

Creative Bot Bulletin #6 is out, featuring Claude 3, Grok-1, and the new book by Ethan Mollick! Read it here: https://mailchi.mp/abf94f2fae0e/creative-bot-bulletin-6-14157421

#generativeAI #LLMs #AI #newsletter

The people of latent.space have built an AI industry news firehose (newsletter, RSS feed) with summaries of the most popular AI subreddits, Twitter accounts and Discord servers.

https://buttondown.email/ainews/archive/

#AI #LLM #newsletters #industrynews #generativeAI

Interview with Dwarkesh Patel, a podcaster who interviews domain experts, mostly about technology.

https://meridian.mercury.com/dwarkesh-patel

#podcasts #technology #AI

"(...) the British model builder's extreme infrastructure costs drained its coffers, leaving the biz with just $4 million in reserve by last October. (...)
What's more, it appears that a sizable portion of the cloudy resources Stability AI paid for were being given away to anyone outside the startup interested in experimenting with Stability's models."

https://www.theregister.com/2024/04/03/stability_ai_bills/

#generativeAI #gpu #cloudcomputing #stabilityAI

In the shadow of Silicon Valley

https://www.lrb.co.uk/the-paper/v46/n03/rebecca-solnit/in-the-shadow-of-silicon-valley

> Drawing from an extensive dataset that spans eight platforms over 34 years—from Usenet to contemporary social media—our findings show consistent conversation patterns and user behaviour, irrespective of the platform, topic or time.

https://www.nature.com/articles/s41586-024-07229-y

Beautiful generated art project "Maps for grief" by Louis-André Labadie.

https://louis-andre.net/journal/maps-for-grief-a-breakdown

#generativeart #art

The longer I work in machine learning, the more I feel that AI engineering is just a combination of open source detective activities and bespoke manual dependency resolving.

“Authors are writing these incredible books, and yet when they ask me questions, the thing that keeps them up at night is, ‘How do I create this brand?’” says literary agent Carly Watters.

https://www.vox.com/culture/2024/2/1/24056883/tiktok-self-promotion-artist-career-how-to-build-following

When people are pressured to meet a target value there are three ways they can proceed:

1) They can work to improve the system
2) They can distort the system
3) Or they can distort the data

https://commoncog.com/goodharts-law-not-useful/

#datadriven #data #datascience #business #metrics #businessintelligence

@evawolfangel 👋 Happy to discuss generative AI!

Can we please stop giving feminine names to chatbots?

"Tip: ask Anna one question at a time, and keep it short." :eyeroll:

@koaning just wait until your bank forces you to use an app that is not available for your version of iOS. 🙈

"Stochastic parrot" was included in the shortlist for Word of the Year 2023!

https://americandialect.org/wp-content/uploads/2024/01/2023-Word-of-the-Year-PRESS-RELEASE.pdf

#llms #generativeai #languagemodels #llama #openai #chatgpt

Happy new year, Fedifriends. Here is the coolest Stable Diffusion clip I've seen in a long time.
https://nitter.net/radamar/status/1744987255549354423

#sdturbo #generativeai #imagegeneration #stablediffusion #lora