#LLMss
Ps, this is why LLMss *are different*

Specifically of mechanistic form. Irrespective of implementation or substrate.
November 20, 2026 at 5:51 PM
The notebooks from the three #LLMSS2026 workshops are now public.

Every notebook runs end to end with no network connection and no API key, so you can rerun and adapt them on your own laptop.

talking-to-machines.github.io/llmss-2026-w...
October 30, 2026 at 1:11 PM
That’s a wrap on the inaugural #LLMSS2026. Thank you to City University of Hong Kong, our speakers and everyone who came.

See you at LLMSS 2027!
October 16, 2026 at 3:07 PM
We are open. The inaugural LLMSS conference begins today at City University of Hong Kong, with EPSA support. #LLMSS2026
October 14, 2026 at 5:51 AM
The inaugural LLM and the Social Sciences conference opens today at City University of Hong Kong, with EPSA support.

Ahead of it, organiser Ray Duch (Oxford) on where the field actually stands: synthetic survey respondents compress variance, skew demographically and move when you reword the prompt.
Where we actually are with synthetic respondents
Written on the eve of LLMSS 2026, Hong Kong
talkingtomachines.substack.com
October 14, 2026 at 4:42 AM
They certainly aren’t going to verify by hand. But human-produced math papers aren’t 100% reliable either, and LLMSs have pretty much caught up with us for reliability. As for clarity, they’ve got some catching up to do.
October 7, 2026 at 2:50 AM
Ask the people suggesting that LLMSs are conscious if they think we should be paying them the federal minimum wage for each hour of work or the California minimum wage.
October 3, 2026 at 9:08 PM
I swear that a huge factor behind LLMSs being seen as the future (other than tech companies shoving it unasked for onto every app) is that intelligent people are hoodwinked like they're watching a Victorian parlour trick and turn into credulous morons unwittingly boosting it's limited capabilities
August 13, 2026 at 12:50 PM
From the Cornell Daily Sun: "'AI is, At Its Heart and Best, a Tool’: Students, Faculty Weigh in on the Use of Large Language Models to Read Academic Papers" www.cornellsun.com/article/2026... #LLMSs #GenAI #AI #highered
May 11, 2026 at 1:04 PM
The error could've been because...

(2) multi-modality is still complex, and driven by the model and also how it processes images. As an example, some models are by their nature multi-modal, capable of reading both images and words, whereas some LLMss use an outside model to read images -- (cont'd)
December 22, 2025 at 12:06 AM
Former Heads of Google Travel and Tripadvisor Form AI Startup to Head Off Online Travel Agencies. This is the way … using AI to go direct to consumer or their AI / robot and cut out middlemen, which of course the marketing funnel folks and merchandisers and ad-supported content folks don't want.
Former Heads of Google Travel and Tripadvisor Form AI Startup to Head Off Online Travel Agencies
DirectBooker wants to deliver hotel supply to LLMss, usurping the emerging role of online travel agencies such as Booking.com.
skift.com
July 18, 2025 at 1:07 PM
January 17, 2026 at 5:31 PM
Prompt-based bioinformatics: a new interface for multi-omics analysis. #Bioinformatics #LLMss #PromptBasedBioinformatics #OmicsDataAnalysis @natrevgenet.nature.com #Comment 🧬 🖥️
www.nature.com/articles/s41...
August 30, 2025 at 4:04 PM
Fantastic explainer on LLMss from Timothy B. Lee and Sean Trott.
https://www.understandingai.org/p/large-language-models-explained-with

Highly recommend it if you want something that doesn't assume much technical expertise, but goes well beyond "LLMs just predict the next word".
July 27, 2023 at 7:42 PM
We are seeking candidates who develop and/or apply AI methodologies within the context of clinically relevant neuroscience problems. Candidates may integrate, advance or develop approaches such as NLP, machine learning, computer vision, foundation models, LLMss, and other methods.
LinkedIn
This link will take you to a page that’s not on LinkedIn
lnkd.in
October 17, 2025 at 8:31 PM
So much hyperbole in your claims. Again, LLMs get a ton of press but they are only a small percentage of the actual AI that is impacting our lives. I hate data center pollution too, but LLMs have little to do with that. And avoiding LLMss is still fairly easy, even jf it won’t be soon.
February 12, 2026 at 8:23 PM
Improving Assembly Code Performance with LLMss via Reinforcement Learning
L: https://arxiv.org/abs/2505.11480
C: https://news.ycombinator.com/item?id=44036902
posted on 2025.05.19 at 21:36:28 (c=1, p=3)
May 20, 2025 at 5:04 AM
I will assign this in future classes—it nails one common argument I find unconvincing in regards to students, which is the idea of LLMss as interlocutors to help refine thinking—that only really works if you’re an expert who can spot the mistakes—students are typically not in a position to do that
November 18, 2024 at 5:12 PM
It's not just poisoned LLMSs poisoning the data they go on to cannibalise. Deliberate and intentional poisoning of training data is a vector for hacking the system itself.
June 3, 2024 at 12:48 PM
This is such a good metaphor for LLMss.
This is how I feel three years into Duolingo
June 29, 2024 at 5:38 PM
Y por eso es que los LLMSS no me sorprenden ni nada
November 12, 2025 at 6:19 PM
Anthropomorphism research emphasizes that AI models can “pretend” to have different perceptions during training

Anthropic publishes a new study that finds that artificial intelligence (AI) models can pretend to maintain different perspectives during training while maintaining their original…
Anthropomorphism research emphasizes that AI models can “pretend” to have different perceptions during training
Anthropic publishes a new study that finds that artificial intelligence (AI) models can pretend to maintain different perspectives during training while maintaining their original preferences. On Wednesday, AI companies stressed that this tendency has attracted serious attention because developers will not be able to trust the results of security training, a key tool to keep large language models (LLMSs) away from harmful behavior.
techwordnews.com
February 23, 2025 at 4:29 AM
Google Titans AI architecture unveils the ability to solve long-term memory problems in AI models

Google researchers last week launched a new artificial intelligence (AI) architecture that allows large language models (LLMSs) to remember the long-term context of events and topics. The mountain…
Google Titans AI architecture unveils the ability to solve long-term memory problems in AI models
Google researchers last week launched a new artificial intelligence (AI) architecture that allows large language models (LLMSs) to remember the long-term context of events and topics. The mountain view-based tech giant published a paper on the topic, where researchers claim that AI models trained using the architecture show more "human-like" memory retention capabilities. It is worth noting that Google abandoned traditional transformer and recurrent neural network (RNN) architectures to develop a new approach to teaching AI models how to remember context information.
techwordnews.com
February 16, 2025 at 12:53 PM
Google Titans AI architecture unveils the ability to solve long-term memory problems in AI models

Google researchers last week launched a new artificial intelligence (AI) architecture that allows large language models (LLMSs) to remember the long-term context of events and topics. The mountain…
Google Titans AI architecture unveils the ability to solve long-term memory problems in AI models
Google researchers last week launched a new artificial intelligence (AI) architecture that allows large language models (LLMSs) to remember the long-term context of events and topics. The mountain view-based tech giant published a paper on the topic, where researchers claim that AI models trained using the architecture show more "human-like" memory retention capabilities. It is worth noting that Google abandoned traditional transformer and recurrent neural network (RNN) architectures to develop a new approach to teaching AI models how to remember context information.
techwordnews.com
February 12, 2025 at 6:53 AM