Everyone wants to be data driven, but no one wants to let data drive. Human capitalist.
Denver, CO
Joined November 2011
- Tweets11.4K
- Following1.1K
- Followers464
- Likes23K
Kevin Hill retweeted
Questions around AI are often framed as a battle between man and machine, ignoring what's possible when they team up instead.
TED devoted a session highlighting ways that cutting-edge artists are engaging with AI.
trib.al/3QN9EiF
Kevin Hill retweeted
🔥Are you an ML scientist interested in connectomics? Become part of our amazing team of scientists and engineers and join our efforts in creating the next generation of brain mapping technology!🔥 (1/3)
Kevin Hill retweeted
What can we learn from “the rare humble bot,” an AI programmed to be awful at chess? It’s a “reminder that intelligent programs can still surprise, delight, and even teach within a clear framework that the user controls,” writes @_jacobsweet. theatlantic.com/technology/a…
Kevin Hill retweeted
Attackers are using artificial intelligence (AI) to enhance conversational scams, such as the so-called "pig butchering" social engineering scams, over mobile devices, @proofpoint researchers say. #cybersecurity #infosec #ITsecurity bit.ly/3A7iTsr
Kevin Hill retweeted
I’ve been reluctant to try ChatGPT. Today I got over that reluctance. Now I understand why I was reluctant.
The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x. I need to recalibrate.
Kevin Hill retweeted
The game theory of prompt injections marginalrevolution.com/margi…
Kevin Hill retweeted
Sparse Pre-training and Dense Fine-tuning for LLMs -- a 2.5x reduction in pre-training FLOPs, without a significant loss in accuracy on the downstream tasks relative to the dense baseline.
arxiv.org/abs/2303.10464
Nice! We studied GPT-3 pre-training and fine-tuning in our recent work, where we show how you can get 60% FLOPs savings during pre-training using weight sparsity and still preserve downstream accuracy arxiv.org/abs/2303.10464
Kevin Hill retweeted
The Open Assistant chat corpus just dropped, 100k-scale chat/instruction dataset from thousands of participants. The era of open data is upon us.
High quality metadata, incl. toxicity scores, attached to each record.
huggingface.co/datasets/Open…
Kevin Hill retweeted
What have the eight ex-Google Brain authors of the Transformers paper been doing since December 2017? 🧐
Let’s find out together! 🧵
Paper: arxiv.org/pdf/1706.03762.pdf
Sources: Google, Pitchbook, LinkedIn
// @ashVaswani @NoamShazeer @aidangomezzz @nikiparmar09 @YesThisIsLion @kyosu
@lukaszkaiser @ilblackdragon
Kevin Hill retweeted
With the Dolly 2.0 release yesterday we heard a lot of people saying they wanted smaller models that can fit on consumer grade GPU’s.
Here you go! Checkpoints from 2.8B and 6.9B now on the Databricks Hugging Face page! 🐑🚀
huggingface.co/databricks/do…
Kevin Hill retweeted
Here's all of biomedical research.
Each dot is one of 21 million papers, based on connections between their abstracts identified by AI. Scroll for the tour, which is both beautiful & shows the method's ability to answer important questions about science. static.nomic.ai/pubmed.html
Kevin Hill retweeted
ok, I think I just had a mini prompting breakthrough
These images are unedited, straight out of Midjourney v5. I'll explain what I did.
🧵Thread w/ some results + workflow
Kevin Hill retweeted
Every Bro who paid $20 for GPT-4:
The Prompt:
Act as: Editor
Degree of revision: Substantial Revision
Type of edit: Enhance clarity and consistency
Change style to: Academic, PhD Work
Change tone to: Analytical
Change reader comprehension level to: advanced, assume extensive prior knowledge
Change length to: 1000 Words
My Text:
Brav.. hav u chaked the gpt-4 thing?? U can writ any branles drunken drivel in it and it changes it to any level even PhD hyper brain style.. truili insanse.. you, got to check it bro! Its a tru gift for all of us iliterate assholes around the the global flat plane! Now we can have universit jobs.. hhah crazi stuff...
Kevin Hill retweeted
Whoa.. still not convinced of AI Agents? This might change your mind...
I pretended to be a fake shoe company and gave AutoGPT a simple objective:
- Do market research for waterproof shoes
- Get the top 5 competitors and give me a report of their pros & cons
Here's how it went:
Kevin Hill retweeted
I had to try it. This video was made without a camera or VFX knowledge.
I generated the clip in Runway #gen2 with the text prompt "a symmetrical man being interviewed at a bar"
Then I brought it into @WonderDynamics and dragged the alien over the actor in the scene and hit render.
Imagine this workflow in the future....
Kevin Hill retweeted
Bold proposal from @tylercowen: perhaps folks who are good at training neural networks with stochastic gradient descent are not, in fact, also the best people to figure out how to manage a complex and risky technology transition...
marginalrevolution.com/margi…
Kevin Hill retweeted
Lots of folks misunderstanding how GPT 3.5 & 4 work, such as in the take below.
"Predict text" is the *pre-training* objective, but it's not what's used for the final model. The final model's objective is, in fact, very similar to adversarial loss (i.e similar to a GAN).
Kevin Hill retweeted
Accurate inflation monitoring requires tracking a broad range of goods beyond food items, which poses challenges in data-poor areas. This @WorldBank study uses innovative techniques to construct multiple machine-learning models to predict missing data. wrld.bg/wYmh50NzNzW
Kevin Hill retweeted
Announcing #PromptHacks23 🚀🚀🚀
Online GPT-4 hackathon with a $10,000 USD cash prize pool taking place April 28-30!
Participants of all skill levels are welcome - we have 250 spots available. Judges from @OpenAI, @Intuit , @IndexVentures and @voiceflow !
Link in thread 👇
