I have a thought about an easier way to build NNMs. What if I took a topic model and created embeddings for, say, every sentence in the Gutenberg collection and the English Wikipedia (as a start). Then ran the word2vec algorithm on those embeddings in sequence? I think that I should get a new embedding space that has the sequential relationships between topics that should be able to accommodate trajectories. This could be validated by drawing trajectories on a UMAP reduced representation of the data. I think.

And boy are there a lot of embedding models

Tasks

• BSO – done
• LLC – done
• Send new text to V – done
• CA24154 – Networking European Security Knowledge

SBIRS

• Change AWS password – done
• 9:00 standup – done
• 3:30 ODIN? For some meeting tomorrow?
• 4:00 SEG – done. Good progress.

Inside the AI Revolution: A Two-Day Odyssey at École Polytechnique

• The article is annoying AI writing, but if you can get past that, the content is pretty good

• Oddly, in some ways I disagree with this. It is the hallucinatory elements of LLMs that are the most interesting.

A language model built for the public good

• ETH Zurich and EPFL will release a large language model (LLM) developed on public infrastructure. Trained on the “Alps” supercomputer at the Swiss National Supercomputing Centre (CSCS), the new LLM marks a milestone in open-source AI and multilingual excellence.

Apertus: A fully open, transparent, multilingual language model

• EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) released Apertus today, Switzerland’s first large-scale, open, multilingual language model — a milestone in generative AI for transparency and diversity.
• The model on HF

Tasks

• Pinged Dr. Veronica Riniolo, and she responded that I need to apply for the activity. So:
  • Set up COST account – done
  • Finish application. I think I need to bundle all the relevant abstracts and use them for the 150 word descriptions. Done! That took a while
• Groceries – return the fuzzy tomatoes, drop off the plastic, and swing by TJ
• Rolling in V’s edits – done!
• Need to see about LLC tomorrow

SBIRs

• Roll in text as it becomes available
• Start looking at GPT-2 layers again. I want to make animations first

It’s September, and I think we can say goodbye (for now?) to HazyHot&Humid.

How Elon Musk Is Remaking Grok in His Image – The New York Times

Non-zero-sum games

• Hi, I’m Non-Zero-Sum James, your companion on this exploration of win-win games and how they are essential for a better future. Each week we’ll explore a new aspect of game theory, moral philosophy, ethical economics and artificial intelligence—looking to solve the complex problems we face in our world together.

This looks really interesting: AIces 2026 – 1st INTERNATIONAL SCHOOL ON THE COGNITIVE, ETHICAL AND SOCIETAL DIMENSIONS OF ARTIFICIAL INTELLIGENCE

• The event will have a global scope along 3 thematic lines: cognition, ethics, and society. It will cover current debates about: AI and philosophy of mind; cognitive architectures; machine learning and cognitive development; large language models and visual information; robotics and embodied cognition; neuroscience-inspired AI; algorithmic bias and fairness; transparency and explainability; accountability and responsibility; privacy and surveillance; autonomy and control; AI impact on human values and social inequalities; the future of work and automation; governance, regulation and public policies; AI, human rights and democracy; AI and global development; information and AI education.

Tasks

• Groceries – done, but I accidentally bought some fuzzy tomatoes
• Storage – done. Sigh
• Get rolling on CA24150 – done!
• Roll in more edits
• Add some quotes, and give Democracy, a Reader a shout out in the quotation section.
• Write a note in the comments on the ICTAI AI(?) review

SBIRs

• 9:00 Standup – done
• Work on the Q2 report – done, except for the communication section and Matt’s part on software reviews
• Looks like I still have a lot of training to do

GPT-5 prompting guide

• We’ve seen significant gains from applying these best practices and adopting our canonical tools whenever possible, and we hope that this guide, along with the prompt optimizer tool we’ve built, will serve as a launchpad for your use of GPT-5. But, as always, remember that prompting is not a one-size-fits-all exercise – we encourage you to run experiments and iterate on the foundation offered here to find the best solution for your problem.

Contrastive Representations for Temporal Reasoning

• In classical AI, perception relies on learning spatial representations, while planning—temporal reasoning over action sequences—is typically achieved through search. We study whether such reasoning can instead emerge from representations that capture both spatial and temporal structure. We show that standard temporal contrastive learning, despite its popularity, often fails to capture temporal structure, due to reliance on spurious features. To address this, we introduce Contrastive Representations for Temporal Reasoning (CRTR), a method that uses a negative sampling scheme to provably remove these spurious features and facilitate temporal reasoning. CRTR achieves strong results on domains with complex temporal structure, such as Sokoban and Rubik’s Cube. In particular, for the Rubik’s Cube, CRTR learns representations that generalize across all initial states and allow solving the puzzle much faster than BestFS—though with longer solutions. To our knowledge, this is the first demonstration of efficiently solving arbitrary Cube states using only learned representations, without hand-crafted search heuristics.

Tasks

• 2FA – done
• Bills – done
• Chores – done
• Dishes – done
• Water plants – done
• 11:30 Atwaters – done
• 3:30 – to DC – well, kind of. I was running late, but had the bike all loaded up and was just about to get on to I95 when I realized that I’d forgotten my helmet. Turned around and went home

Setting up the COST proposal project (CA24150 – participation instructions). Need to reach out to a lot of people to build the team.

Send note to No Starch about the P&P. 80 tickets at $15.00-ish for a Tuesday evening event. Done

College students are bombarded by misinformation, so this professor taught them fact-checking 101 − here’s what happened

• For Generation Z, born between 1997 and 2012, social media – especially YouTube, TikTok, Instagram and Snapchat – has become their source of information about the world, eclipsing traditional news outlets. In a survey of more than 1,000 young people ages 13 to 18, 8 in 10 said they encounter conspiracy theories in their social media feeds each week, yet only 39% reported receiving instruction in evaluating the claims they saw there.
• We narrowed the focus of our program to skills essential to being an informed citizen, such as “lateral reading” − that is, using the full context of the internet to judge the quality of a claim, identify the people or organizations behind it and assess their credibility. Rather than fixate solely on the message, we taught students to vet the messenger: What organizations stand behind the claim? Does the source of the claim have a conflict of interest? What are the source’s credentials or expertise?

SBIRs

• 9:00 standup – done
• Put in a bunch of text about MORS WG30 activities at their annual symposium in the BP. Need to add a bit more just to mention the typical number of sessions at these events – done. Seems good enough to send along
• 4:00 SEG meeting quick. Everything is on schedule
• Did laptop return things

[2506.02153v1] Small Language Models are the Future of Agentic AI

• Large language models (LLMs) are often praised for exhibiting near-human performance on a wide range of tasks and valued for their ability to hold a general conversation. The rise of agentic AI systems is, however, ushering in a mass of applications in which language models perform a small number of specialized tasks repetitively and with little variation.
  Here we lay out the position that small language models (SLMs) are sufficiently powerful, inherently more suitable, and necessarily more economical for many invocations in agentic systems, and are therefore the future of agentic AI. Our argumentation is grounded in the current level of capabilities exhibited by SLMs, the common architectures of agentic systems, and the economy of LM deployment. We further argue that in situations where general-purpose conversational abilities are essential, heterogeneous agentic systems (i.e., agents invoking multiple different models) are the natural choice. We discuss the potential barriers for the adoption of SLMs in agentic systems and outline a general LLM-to-SLM agent conversion algorithm.
  Our position, formulated as a value statement, highlights the significance of the operational and economic impact even a partial shift from LLMs to SLMs is to have on the AI agent industry. We aim to stimulate the discussion on the effective use of AI resources and hope to advance the efforts to lower the costs of AI of the present day. Calling for both contributions to and critique of our position, we commit to publishing all such correspondence at this https URL.

Getting back into the academic proposal swing. Seeing what’s available

SBIRs

• 10:00 Emerson meeting
• 4:00 ADS tagup

From https://bsky.app/profile/segyges.bsky.social/post/3lxclvdliks2p

npj Complexity is an open access, international, peer-reviewed journal dedicated to publishing the highest quality research on complex systems and their emergent behavior at multiple scales. 

Tasks

• Profs and Pints tonight! Get there 45 minutes early and bring the laptop
• Start updating the calls section on the Cognitive Commons template. See what makes sense for the first submission
• New bushes today?

SBIRs

• 9:00 Sprint planning. Done
• Start on Q2 report. Started. The changed the template to the revised (and as yet not approved in writing) SOW, and created a Q2_text folder with the blanks
  • Reworking the tasking – done

Had an annoying morning trying to get VFS to work properly

Tasks

• Slides and timings for P&P talk – done? War Room takes 8 minutes to read, so I have about 22 minutes to talk. Good discussion with Jimmy to work out final details
• Dead shrubs? Shrubs are gone, but replacements are not in yet
• Powerwashing quote? Soon

SBIRs

• 9:00 Sprint demos – done
• 12:30 Survey review – I dunno.
• 3:00 Sprint planning – two stories: quarterly status report and the AW proposal ROI section – delayed

September is closing in fast

From Mastodon

Tasks

• Read this page on associationism
• Send Vanessa new vignette – done
• Bills – done
• Chores – done
• Mow – done
• Weed – done
• Laundry – mostly done
• Vacuum and organize shop – done for now
• Dishes – done
• Research storage
• Metal to Aaron – done
• Read through stories for P&P and time them – first pass and editing

SBIRs

• Write up notes from yesterday and see when the next report is due. It will have the new tasks and numbering – Done

Saw my first AI review today. Pretty sad:

Tasks

• Start proposal 14 review – done! I HAVE NO MORE REVIEWS TO DO!
• Submit No Starch Press proposal – DONE! Need to look for other possible publishers though. Still, it’s nice to get this far.
• Send Vanessa new vignette
• Research storage

SBIRs

• 9:00 standup. See if Ron needs some tag-teaming – not yet
• 4:00 SEG meeting – need to write up the notes. Basically, Ron and John are going to try and get interprocess communication going.
• Work on the A3IW proposal – finished! Well a first pass anyway. I thing a “potential market” section might help
• Circle around with Aaron to see if we should really write a WH/AI proposal – pinged

WtaF? ‘I Want to Try and Get to Heaven’: Trump Gets Reflective on ‘Fox & Friends’

Tasks

• There is some back and forth on the review of proposal 7. It may just be a no-cost extension, which is an easy “yes.” Ok, got some direction. I need to rewrite and submit – done
• Submit proposal 13 review – done
• Start proposal 14 review – tomorrow
• Work on No Starch Press proposal – done. I’ll send it in tomorrow
• Roll in KA edits – done!
• Research storage

SBIRs

• Nothing on the schedule today. See if Ron can get results to share on Thursday – not really. He didn’t look up solutions and wrote himself into a corner. Going to let him work himself out of it. He really needs to stop doing this
• Work on the A3IW proposal – nope
• Circle around with Aaron to see if we should really write a WH/AI proposal – pinged

Pinged Carlos back.

No Starch Press pinged back! I need to put together a proposal after finishing the reviews

Tasks

• Review proposal 13 – done. May need to revisit 7, but I’m confused for now
• Kyle today? Yup, the torch, the brake, and the endmill are gone. The lathe is going to take some more works – it’s a bit over 1,000 lbs

SBIRs

• 9:00 standup
• Work with Aaron on the white paper – started. It’s going to be more of a business proposal. I think I’m going to run it through Gemini to see if it can come up with a good framework because frankly, I just don’t want to do any more BD that won’t go anywhere.

President Trump’s War on “Woke AI” Is a Civil Liberties Nightmare | Electronic Frontier Foundation

• The White House’s recently-unveiled “AI Action Plan” wages war on so-called “woke AI”—including large language models (LLMs) that provide information inconsistent with the administration’s views on climate change, gender, and other issues. It also targets measures designed to mitigate the generation of racial and gender biased content and even hate speech. The reproduction of this bias is a pernicious problem that AI developers have struggled to solve for over a decade.
• A new executive order called “Preventing Woke AI in the Federal Government,” released alongside the AI Action Plan, seeks to strong-arm AI companies into modifying their models to conform with the Trump Administration’s ideological agenda.

Russia is quietly churning out fake content posing as US news – POLITICO

• According to misinformation tracker NewsGuard, the campaign — which has been tracked by Microsoft’s Threat Analysis Center as Storm-1679 since at least 2022 — takes advantage of high-profile events to pump out fabricated content from various publications, including ABC News, BBC and most recently POLITICO.
• “They are just throwing spaghetti, trying to see what’s going to stick on a wall,” said Ivana Stradner, a researcher on Russia at the Foundation for Defense of Democracies, a Washington think tank.

Build a Small Language Model (SLM) From Scratch

Tasks

• Write review for proposal 7 – done. That was a lift.
• Machine shop pickup? Tomorrow
• Work on rolling in edits. Finished the War Room story
• Put together a proposal for No Starch Press

SBIRS

• Work with Ron a bit?
• Overleaf doc A3IW

Great ride yesterday, even if we did get rained on in the middle

Tasks

• Switch UPS – done
• Bills – done
• Chores – done
• Dishes – done
• Schedule power wash – started
• Read Proposal 14 – done
• Ping Nathan – done
• Wash truck – done. Shiny!
• Picked up the recumbent
• Put Peter Turchin in P33 somewhere. Maybe simulation?
• Ping Carlos to see if there will be a recording of Trust in human-technology teams: from decision-support to Generative AI – done