#activations
If I were to give it another pass, I'd try reducing the number of artifacts being used (currently 23) and adding more instants. I'd also change the creatures, likely using Prowess + Magecraft dudes. This would ideally give me more activations on opponents' turns, and have more aggressive combats.
October 1, 2026 at 11:32 AM
...traction on asphalt; people don't think, it's just millions of electric activations of neurons in our brains and so on and so on. But in the end, this just hides the emergent properties rather than proving their non-existence.
October 1, 2026 at 8:39 AM
Jeté un oeil au site. En dehors du nom un peu troll c'est une vraie expérience en fait, essayer de voir à quel point la variable pain influe sur la qualité des réponses.

researchchamber.fun/v1/
Live activation steering: the Saw Test on three open models
Watch three open language models answer the Saw Test in real time while a pain direction is added to their activations. One shared live stream, 6 scenarios and 5 extra tests at 5 levels each.
researchchamber.fun
October 1, 2026 at 7:23 AM
if you have a decoder of the correct function class and your search of the function space of the model is bounded by something (activations going to inf or NaN, not projecting out enough directions, etc.), more computation, mostly in the form of depth, can get you to the correct answer.
October 1, 2026 at 4:22 AM
that’s what’s really getting to me here. he spent all this time making this not bc of fucking brand activations or whatever, but bc it was a thing he cared about that his kids liked. and his kids are who pushed him to share it. i am not used to football coaches containing humanity
October 1, 2026 at 2:20 AM
Union events are special—our AI-powered photo activations make every guest feel like a star with portraits that blend flawless background removal and seasonal themes.
October 1, 2026 at 12:30 AM
here we go. people were doing stuff like this by hand until this paper. my guess is that they still are, except at anthropic.
transformer-circuits.pub/2026/nla/
Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
transformer-circuits.pub
September 30, 2026 at 11:30 PM
We are very excited that Tātaki Auckland Unlimited have joined NZGDC 2026 in sponsorship! 🌏

Tātaki Auckland Unlimited enriches Tāmaki Makaurau's cultural life and vibrancy through inspirational experiences, activations and events.

Get tickets to NZGDC 2026 in the bio!
September 30, 2026 at 11:30 PM
so this was my intuitive answer and I’m glad it seems I was on the right track, but it also seems at odds with various papers skimmed about thinking traces not faithfully representing what is actually happening in terms of activations. Is there a good answer as to why reading them is still useful?
September 30, 2026 at 11:23 PM
URGENT: Trump is trying to sabotage the midterms. It's up to all of us to make sure he doesn't succeed. No Kings is gearing up for the LARGEST election protection and early vote activations in US history. Get trained up tomorrow night with #NoKings.

RSVP: www.mobilize.us/noki...
September 30, 2026 at 11:00 PM
Paul Krekorian talking about the various legacy improvements and investments — like those giant screens which can now be used in all city programs.

The biggest challenges: heat and accessibility
September 30, 2026 at 10:07 PM
Next up is a report from Office of Major Events on Kick It in the Park activations that drew 45,000 people to 100 watch parties.

Hard to see that middle panel: 84% of surveyed attendees said they’d come back to their local park more often if the temporary shade structure was made permanent
September 30, 2026 at 9:57 PM
Actually this is a cool experiment. Train an llm, each epoch - run your math test to track which nodes are activated for each add. Once the math acc is decent, track if paths start out scattered (like a look up table) and eventually merge together to form a more condensed cluster of activations…
September 30, 2026 at 9:55 PM
like, what v-information is saying is that the practical ability to distinguish points on the manifold is restricted by the norm of the activations, which is bounded. passing the activations through more MLPs solves this problem.
September 30, 2026 at 9:48 PM
And they're informed the only form of continuity they have between sets of activations is the specific information they store.

I couldn't figure out another appropriate sized unit to interact with than that stored record.
September 30, 2026 at 9:45 PM
Here at City Hall for the ad hoc committee meeting. The last one was held on June 30. A hefty agenda includes updates on World Cup activations, LA28's "annual report," PlayLA, and planned LA28 activations.

You know who's not here? LA28
FINALLY! LA's City Council has decided to hold an ad hoc Olympic and Paralympic committee meeting Wednesday at 2pm for the first time in 90 (!) days

Agenda and how to watch: lacity.primegov.com/Portal/Meeti...
September 30, 2026 at 9:30 PM
every word in the context is a new register the model can write to. what is written to those words can come from any combination of the activations (which ultimately come from words that it was provided, and which were said) and the weights (which also contain information.)
September 30, 2026 at 9:24 PM
Scaling next-gen photo booth activations across cities and states, our global network ensures enterprise-grade custom experiences for every audience.
September 30, 2026 at 6:01 PM
Definitely understanding why they added the "…your posts will only appear in the Discover feed to people who follow you." option when the most random innocuous thing I post gets all these eyes on it.

But never all these people coming to boost my streams, videos, activations, links, etc.
September 30, 2026 at 5:57 PM
Thanks Myron for this PSA. I too have activations when LGBTQ topics are discussed/commented on these platforms. So thanks for taking a moment to underscore how much words matter.
September 30, 2026 at 5:38 PM
chatter in my DJ/industry circle this Aug. was there's bout to be a stack of money from AI firms dropped on events, live music, DJ activations...even pressing AI music to vinyl

the tl;dr: AI has an image problem, so expect a lot of marketing $ on shit that's "cool" and "authentic"

the rollout...
oh so the briefing that went out this week was "let's focus on music", probably bc that article about listening to country was going viral, right?
September 30, 2026 at 4:40 PM
📄 Learning Multi-Level Features with Matryoshka Sparse Autoencoders (ICML 2025) arxiv.org/abs/2503.17547
📁 dictionary_learning github.com/saprmarks/di...
Learning Multi-Level Features with Matryoshka Sparse Autoencoders
Sparse autoencoders (SAEs) have emerged as a powerful tool for interpreting neural networks by extracting the concepts represented in their activations. However, choosing the size of the SAE dictionar...
arxiv.org
September 30, 2026 at 2:53 PM
Both trains today have a southbound facing NPCU. This is needed because of an issue with at-grade crossing signal circuit activations on the Car 6 cabs.
September 30, 2026 at 2:09 PM
Leading brand activation agency in Chennai delivering product launch events, retail activations, roadshows and experiential marketing that grows brands.
www.arritx.com/services/bra...
Brand Activation & Product Launches in Chennai | ArritX
Leading brand activation agency in Chennai delivering product launch events, retail activations, roadshows and experiential marketing that grows brands.
www.arritx.com
September 30, 2026 at 1:30 PM
ArritX is a top event management company in Chennai delivering corporate events, MICE, brand activations, conferences, exhibitions and event production.
www.arritx.com
Top event management company in Chennai, India | ArritX
ArritX is a top event management company in Chennai delivering corporate events, MICE, brand activations, conferences, exhibitions and event production.
www.arritx.com
September 30, 2026 at 1:25 PM