/
Navigation
C
Chronicles
Browse all articles
C
E
Explore
Semantic exploration
E
R
Research
Entity momentum
R
N
Nexus
Correlations & relationships
N
~
Story Arc
Topic evolution
S
Drift Map
Semantic trajectory animation
D
P
Posts
Analysis & commentary
P
Browse
@
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
?
Concept Search
Semantic similarity search
!
High Impact Stories
Top coverage by position
+
Sentiment Analysis
Positive/negative coverage
*
Anomaly Detection
Unusual coverage patterns
Analysis
vs
Rivalry Report
Compare two entities head-to-head
/\
Semantic Pivots
Narrative discontinuities
!!
Crisis Response
Event recovery patterns
Connected
Nav: C E R N
Search: /
Command: ⌘K
Embeddings: large
VOICE ARCHIVE

@blancheminerva

@blancheminerva
10 posts
2025-11-30
DeepSeek is more aligned to my values than any US frontier lab. I wish EleutherAI had the billions it took to do this work, but since we have less than 1/1000th of that I take a lot of joy in seeing someone else do what needs to be done.
2025-11-30 View on X
The Decoder

DeepSeek says its new DeepSeekMath-V2 model got gold-medal level status on the International Mathematical Olympiad 2025 and Chinese Mathematical Olympiad 2024

where models prove formal mathematical theorems—GPT-5 scores 20%.  Gemini Deep Think IMO Gold hits 65.7%.  DeepSeek Math V2 (Heavy) scores 61.9%.  That's second place—but Gemini is...

2025-11-29
DeepSeek is more aligned to my values than any US frontier lab. I wish EleutherAI had the billions it took to do this work, but since we have less than 1/1000th of that I take a lot of joy in seeing someone else do what needs to be done.
2025-11-29 View on X
The Decoder

DeepSeek says its new DeepSeekMath-V2 model got gold-medal level status on the International Mathematical Olympiad 2025 and Chinese Mathematical Olympiad 2024

Chinese startup Deepseek reports its new DeepseekMath-V2 model has reached gold medal status at the Math Olympiad …

2024-01-25
In today's research landscape broadening access to large-scale computing resources and HPC expertise is essential for advancing AI research and equalizing opportunities across institutions. @AiEleuther is thrilled to partner with @NSF to make AI at scale more accessible.
2024-01-25 View on X
Forbes

The US National Science Foundation partners with OpenAI, Meta, Nvidia, Microsoft, and others to launch NAIRR, a pilot to expand researchers' access to AI tools

combining the best of safety, privacy, & use! We're hiring!!! https://openmined.org/... @whostp : Today, @NSF launched the National AI Research Resource pilot. Giving researchers, ...

2023-12-05
Nothing highlights just how fragile alleged “safety filters” are than monkey-patching ToS to avoid known issues. Any conversation about the safety of LLMs that doesn't acknowledge that current “SOTA” safety filters don't prevent bad people from doing bad things is disingenuous.
2023-12-05 View on X
404 Media

Asking ChatGPT to repeat words “forever” now violates OpenAI's terms of service, after Google researchers showed that doing so led to revealing training data

2023-04-20
Really great work! Doing analysis like this is exhausting and time consuming, but very important. There are some really good documentation ideas here that I didn't think about with the Pile, and hope to do in the future. https://twitter.com/...
2023-04-20 View on X
Washington Post

Analysis: Google's C4 dataset, used to train LLMs like Meta's LLaMA, has troubling content from 4chan, Kiwi Farms, white supremacist site Stormfront, and more

AI chatbots have exploded in popularity over the past four months, stunning the public with their awesome abilities …

2023-03-29
@CerebrasSystems You can also download the models yourself from HuggingFace, and they should be compatible with @AiEleuther's LM Eval Harness right out of the box: https://huggingface.co/...
2023-03-29 View on X
SiliconANGLE

Cerebras open sources seven GPT-based LLMs, ranging from 111M to 13B parameters and trained using its Andromeda supercomputer for AI, on GitHub and Hugging Face

Artificial intelligence chipmaker Cerebras Systems Inc. today announced it has trained and now released seven GPT-based large language models …

2023-03-22
Phenomenal piece by @random_walker, with an extremely interesting anecdote about automation: Just one of the 270 jobs in the 1950 census has been eliminated by automation: elevator operator. https://twitter.com/...
2023-03-22 View on X
AI Snake Oil

OpenAI may have tested GPT-4 on its training data, violating the cardinal rule of ML, and GPT-4's exam performance says little about its real-world usefulness

OpenAI may have tested on the training data.  Besides, human benchmarks are meaningless for bots.

2022-08-31
@simonw Whats the reason for using a sample instead of the whole thing? Compute limitations? Too much results for a human to sift through? Something else? If you just need more compute I would be happy to help by providing however much you need.
2022-08-31 View on X
Waxy.org

A look at 12M+ of the 2.3B images used to train Stable Diffusion reveals ~47% came from only 100 domains; the largest number, or 8.5%, originated from Pinterest

laion-aesthetic-datasette Use Datasette to explore … Tweets: Andy Baio / @waxpancake : What images are in the massive dataset that trained the Stable Diffusion text-to-image AI mod...

2021-09-20
@Facebook has taken courses from @GoogleAI in gaslighting it seems. @WSJ is hardly the first to document the wide-spread, sanctioned, and systematic abuse of moderation powers at Facebook, but it seems there's always a deeper rabbithole. https://www.wsj.com/... https://twitter.com/...
2021-09-20 View on X
About Facebook

Facebook says WSJ's series contained deliberate mischaracterizations and conferred egregiously false motives to Facebook's leadership and employees

and burying the findings Charlie Osborne / ZDNet : Facebook rebukes WSJ over investigation on the platform's ability to harm, ‘toxic’ impact Parmy Olson / Bloomberg : Facebook's Be...

2021-09-11
The people at @CohereAI are cool, but anyone who thinks that the way you “democratize” something is to start a start-up to sell it needs a new dictionary. https://www.fastcompany.com/ ...
2021-09-11 View on X
Fast Company

Toronto-based Cohere, founded by two ex-Google Brain team members to commercialize natural language AI models, raises $40M Series A led by Index Ventures

Mark Sullivan / Fast Company : Tweets: @blancheminerva , @jordanjacobs10 , @jordanjacobs10 , @nathanbenaich , and @nickfrosst . Thanks: @jenholmes Tweets: Stella Rose Biderman / @...