carla retweeted
Replying to @jacobrintamaki
I’m not even joking a little bit when I say whoever runs, Chick-fil-A needs to be running more things
These are my Lebanese and Korean grandfathers, from opposite sides of the world, holding their grandson. Last week, that grandson made history as the first Asian American and Arab American elected to represent Massachusetts in Congress.
America, the beautiful melting pot. ❤️🇺🇸
It's 2018 again.
Everyone is excited about all these crazy images generated from GANs.
The same guy that made GANs just came out with a deep learning textbook with Bengio.
NIPS just changed their name to NeurIPS and you wonder if that's actually going to stick.
You train your Encoder-Decoder LSTM model with attention.
Some crazy people are starting to use this new model architecture called a Transformer.
Other crazy people are starting to fine-tune with something called REINFORCE to maximize BLEU and ROUGE scores.
OpenAI just solved a Rubik's cube with a robot hand.
Life is good.
carla retweeted
Why is everything a chart now 📊?
X's algorithm weights sharing a post via dm 7.1x as heavily as likes and copy-link sharing 28.2x. On Oct 26th, Head of Instagram Mosseri reinforced the importance of shares, "sends per reach is one of the most important metrics new creators can track, particularly for reaching non-followers.".
Social media has felt less like social media as we've shifted from a following/friends based feed to the era of interest media. For Instagram, just 7% of time spent on app was viewing friends content in 2025. Most people are watching creators.
So shares are one of the few ways to create social connection and make social media more 'social'. And from a business perspective, they are great at keeping people on platform as the person who receives the content is now back on platform and can get served new ads. Also, the network effect, more friends who share & chat, more reasons you have to keep using it.
Also, for people who have grown up a social-native, sharing content and memes has become an engrained part of communication. It's funny you can 'chat' to people without actually texting but purely communicate via content. I also think there's a deep human need to share and show things with other people, probably some primal/caveman thing there.
So the habit is there, we share content with friends, colleagues, family etc. Since platforms like Slack released in 2014 and started to gamify work comms, the way people shared/interacted on more 'non-work platforms' like Instagram has become pretty much the same behaviour on more 'work' platforms like Slack/Teams. Now compared to 10 years ago, your work comms look much more like social media comms.
As a result, dark social (sharing via private, uncrackable channels like slack/teams/whatsapp/dm) for work is incredibly valuable. Most businesses now have an 'interestingreads' or 'knowledge share' slack/teams channel where people share posts that are helpful or relevant to their industry. If you're a startup and wanting to grow or sell into a company, one anecdotal metric is to show up in that 'interestingreads' over and over again. There are probably 2-3 companies whose content you always end up sharing internally.
I would bet at least a hundred odd businesses shared the Tony Hawk x Ramp collab or Jude Law x Legora campaign in a slack channel with a message like "this is sick" or "super clever" etc. Content that gets shared is usually either for entertainment or education. That's why you're seeing so many companies start to put everything in a graph because it's snackable and educational.
a16z charts of the week is a perfect example of this, every week one of their charts goes viral and gets shared in a bunch of dark social channels. Every business should be hiring a @PeterJ_Walker or @saltzman_jason type who consistently produce high-quality self-owned graphs/charts that always get shares.
Do this =
1.create snackable value-led graphs.
2.creative outside the blob campaigns
carla retweeted
slopsurfing: the art of keeping code at precisely the right level of slop for maximum experimental productivity
carla retweeted
me: "proprietary data"
the pizza CMO i just chatted to:
"proprietary pepperoni"
there's levels to the game
carla retweeted
You can 2x Qwen-0.8b's performance on Math-500 by adding Gaussian noise to the last 4 layers.
RSI is here 🔥🔥🔥 and it's Gaussian!
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
carla retweeted
for many of you who learned DL post LLM era.. you may not know.. but:
--
1) Adam doesnt converge on some convex, 1D functions.
2) Gradient Descent fails in many documented cases.
3) the brain doesnt do backprop. all arguments for this presume at least a cyclic graph, at most a completely different learning procedure that is BP-like, never neurons actually firing backward which is what would be required to actually implement BP.
--
there are other update rules.. many other update rules left to find..
carla retweeted
Stanford Digital Brain Project has launched! We're hoping to make a seriously useful model of the human brain. Please come join us! news.stanford.edu/stories/20…




































