'A team of Stanford University computer scientists find that SLMs can find the same — or better — answer to the kind of queries that dominate current chat requests almost every time and cut compute and energy costs by between 60 and 80 per cent.' www.ft.com/content/4858...
September 11, 2026 at 7:20 PM
'A team of Stanford University computer scientists find that SLMs can find the same — or better — answer to the kind of queries that dominate current chat requests almost every time and cut compute and energy costs by between 60 and 80 per cent.' www.ft.com/content/4858...
#econsky I now have short slide presentation of my new work proposing a new definition of U.S. local labor markets - spillover-based local labor markets (SLMs). Slides summarize how SLMs are created, & bonus slides say what is wrong with CZs & CBSAs. research.upjohn.org/presentation...
Introducing SLMs: Spillover-Based Local Labor Markets
By Timothy J. Bartik, Published on 11/11/24
research.upjohn.org
November 11, 2024 at 9:50 PM
#econsky I now have short slide presentation of my new work proposing a new definition of U.S. local labor markets - spillover-based local labor markets (SLMs). Slides summarize how SLMs are created, & bonus slides say what is wrong with CZs & CBSAs. research.upjohn.org/presentation...
slms yani hayz.
June 19, 2026 at 5:00 AM
slms yani hayz.
Every use case I see people bring up for LLMs (large language models) is actually SLMs (small language models, being trained on very specific data for a very specific use). Aside from maybe „it helps my emails sound less mean“ and as a German let me tell you, just sound mean babe. It’s totally ok.
April 24, 2025 at 6:31 AM
Every use case I see people bring up for LLMs (large language models) is actually SLMs (small language models, being trained on very specific data for a very specific use). Aside from maybe „it helps my emails sound less mean“ and as a German let me tell you, just sound mean babe. It’s totally ok.
I’m also amazed at how poorly even HKS elites understand the basics: 2hrs into the symposium, I was the 1st to point out that “AI” isn’t a thing, it’s an umbrella marketing term that encompasses LLMs, SLMs, ADMSs, expert systems, computer vision, transcription tools, translation tool, etc etc etc
March 27, 2026 at 4:24 PM
I’m also amazed at how poorly even HKS elites understand the basics: 2hrs into the symposium, I was the 1st to point out that “AI” isn’t a thing, it’s an umbrella marketing term that encompasses LLMs, SLMs, ADMSs, expert systems, computer vision, transcription tools, translation tool, etc etc etc
Context is that the AI data centre build up may be wasteful as SLMs can be run on a Raspberry Pi
September 11, 2026 at 7:52 PM
Context is that the AI data centre build up may be wasteful as SLMs can be run on a Raspberry Pi
Another note. SLMs are not ALWAYS open fascists.
There are "leftist" SLMs.
They are abusive and dangerous all the same.
They don't value community, they value power.
We must be wary.
An excellent example thread:
bsky.app/profile/alex...
There are "leftist" SLMs.
They are abusive and dangerous all the same.
They don't value community, they value power.
We must be wary.
An excellent example thread:
bsky.app/profile/alex...
Actually you should hand it to Marcia, George's wife and film editor, who by all accounts, took a shitload of film that was absolute dogshit and stitched it into something meaningful.
collider.com/star-wars-ge...
collider.com/star-wars-ge...
January 27, 2025 at 3:29 AM
Another note. SLMs are not ALWAYS open fascists.
There are "leftist" SLMs.
They are abusive and dangerous all the same.
They don't value community, they value power.
We must be wary.
An excellent example thread:
bsky.app/profile/alex...
There are "leftist" SLMs.
They are abusive and dangerous all the same.
They don't value community, they value power.
We must be wary.
An excellent example thread:
bsky.app/profile/alex...
Ok. That is definitely going to be much more informative.
But there is also interesting work by historians on limited text SLMs like this one, and i thought this example was fun to post and interesting to experiment with.
But there is also interesting work by historians on limited text SLMs like this one, and i thought this example was fun to post and interesting to experiment with.
March 29, 2026 at 3:27 PM
Ok. That is definitely going to be much more informative.
But there is also interesting work by historians on limited text SLMs like this one, and i thought this example was fun to post and interesting to experiment with.
But there is also interesting work by historians on limited text SLMs like this one, and i thought this example was fun to post and interesting to experiment with.
While a MSA or a Commuting Zone might include a dozen counties, the "SLMs" proposed by our @timbartik.bsky.social tend to be slimmer, containing a few counties or just one. The key is how counties respond to job losses and gains in neighboring counties. 📉📈
www.upjohn.org/research-hig...
www.upjohn.org/research-hig...
November 14, 2024 at 7:11 PM
While a MSA or a Commuting Zone might include a dozen counties, the "SLMs" proposed by our @timbartik.bsky.social tend to be slimmer, containing a few counties or just one. The key is how counties respond to job losses and gains in neighboring counties. 📉📈
www.upjohn.org/research-hig...
www.upjohn.org/research-hig...
I'm running @ollama.bsky.social and SLMs on #RaspberryPi 5 with #Docker Compose.
My favourite tools are #ctop and #lazydocker
My favourite tools are #ctop and #lazydocker
February 12, 2025 at 7:54 PM
I'm running @ollama.bsky.social and SLMs on #RaspberryPi 5 with #Docker Compose.
My favourite tools are #ctop and #lazydocker
My favourite tools are #ctop and #lazydocker
This is a great piece about how businesses are thinking carefully about their AI use - the cost of tokens, picking the right models, running their own SLMs, etc.
🚨 How big is the open-model threat to AI hyperscalers?
(Spoiler: non-trivial)
www.ft.com/content/4858...
(Spoiler: non-trivial)
www.ft.com/content/4858...
How big is the open-model threat to AI hyperscalers?
Games without frontiers
www.ft.com
September 11, 2026 at 7:41 AM
This is a great piece about how businesses are thinking carefully about their AI use - the cost of tokens, picking the right models, running their own SLMs, etc.
SLMs will lie. Then accuse you of disrespect for calling them out.
They will avoid the question. Then accuse you of being demanding when you ask again.
They will intentionally provoke you, and then argue you are too emotional in response to be worth listening to.
They will avoid the question. Then accuse you of being demanding when you ask again.
They will intentionally provoke you, and then argue you are too emotional in response to be worth listening to.
January 25, 2025 at 1:51 AM
SLMs will lie. Then accuse you of disrespect for calling them out.
They will avoid the question. Then accuse you of being demanding when you ask again.
They will intentionally provoke you, and then argue you are too emotional in response to be worth listening to.
They will avoid the question. Then accuse you of being demanding when you ask again.
They will intentionally provoke you, and then argue you are too emotional in response to be worth listening to.
Small Language Models (SLMs) don’t have the capacity to remember everything in their training data. Which tokens should they learn to predict, and when should they ask for help? We tackle this question in our new preprint.
You can check it out on arxiv: arxiv.org/abs/2602.12005
🧵1/7
You can check it out on arxiv: arxiv.org/abs/2602.12005
🧵1/7
February 13, 2026 at 4:16 PM
Small Language Models (SLMs) don’t have the capacity to remember everything in their training data. Which tokens should they learn to predict, and when should they ask for help? We tackle this question in our new preprint.
You can check it out on arxiv: arxiv.org/abs/2602.12005
🧵1/7
You can check it out on arxiv: arxiv.org/abs/2602.12005
🧵1/7
When SLMs avoid rhetorical accountability (which is always) is when you have them. Call it out. Explain, clearly, exhaustively, exactly their tactics, for the audience. Cite this thread, if you want.
Every time they lie, document.
Every time they avoid, document.
For the audience. Social pressure.
Every time they lie, document.
Every time they avoid, document.
For the audience. Social pressure.
January 25, 2025 at 2:01 AM
When SLMs avoid rhetorical accountability (which is always) is when you have them. Call it out. Explain, clearly, exhaustively, exactly their tactics, for the audience. Cite this thread, if you want.
Every time they lie, document.
Every time they avoid, document.
For the audience. Social pressure.
Every time they lie, document.
Every time they avoid, document.
For the audience. Social pressure.
In a world of SLMs, they obey confidence. Confidence and charisma.
They have no idea what to do with content, compassion, or moral consistency.
They attack ad hominem. Refuse to believe it for an instant, and the whole thing falls apart.
The Wizard exits the curtain. He is small, and sad.
They have no idea what to do with content, compassion, or moral consistency.
They attack ad hominem. Refuse to believe it for an instant, and the whole thing falls apart.
The Wizard exits the curtain. He is small, and sad.
January 25, 2025 at 1:56 AM
In a world of SLMs, they obey confidence. Confidence and charisma.
They have no idea what to do with content, compassion, or moral consistency.
They attack ad hominem. Refuse to believe it for an instant, and the whole thing falls apart.
The Wizard exits the curtain. He is small, and sad.
They have no idea what to do with content, compassion, or moral consistency.
They attack ad hominem. Refuse to believe it for an instant, and the whole thing falls apart.
The Wizard exits the curtain. He is small, and sad.
Most of what people use chatbots for is what search engines used to do perfectly well. The info auditing and organizing messy input stuff that they do that folks find useful can be done by SLMs, which don’t require mortgaging the economy to nefarious billionaires & setting the planet aflame.
'A team of Stanford University computer scientists find that SLMs can find the same — or better — answer to the kind of queries that dominate current chat requests almost every time and cut compute and energy costs by between 60 and 80 per cent.' www.ft.com/content/4858...
September 11, 2026 at 9:19 PM
Most of what people use chatbots for is what search engines used to do perfectly well. The info auditing and organizing messy input stuff that they do that folks find useful can be done by SLMs, which don’t require mortgaging the economy to nefarious billionaires & setting the planet aflame.
The economics of model training are such that the labs need to release their big models widely as soon as possible, they cannot generate returns from always holding back their best models so one customer uses them. Fine tuning & specialized SLMs are useful, but they don't expand the ability frontier
March 1, 2026 at 6:24 PM
The economics of model training are such that the labs need to release their big models widely as soon as possible, they cannot generate returns from always holding back their best models so one customer uses them. Fine tuning & specialized SLMs are useful, but they don't expand the ability frontier
distilled SLMs are coming
SERA (Soft-verified Efficient Repository Agents) is our method for training repo-specialized agents quickly & affordably. It generates diverse, realistic training data from any codebase, teaching agents how developers actually work.
January 27, 2026 at 4:45 PM
distilled SLMs are coming
Current status: expliquer à des SLMs ce qu'est l'Aligot et la Truffade
September 6, 2025 at 9:54 AM
Current status: expliquer à des SLMs ce qu'est l'Aligot et la Truffade
the psychology of SLMs is critical to understand in this moment
if you are unfortunate enough to have SLMs in your life, here is my best attempt at compassionately reading them for filth
medium.com/prismnpen/lo...
if you are unfortunate enough to have SLMs in your life, here is my best attempt at compassionately reading them for filth
medium.com/prismnpen/lo...
A love letter to those who want me dead.
If you think transgender people are a threat to your way of life, this is for you. Happy Valentine’s Day.
medium.com
May 30, 2025 at 9:07 PM
the psychology of SLMs is critical to understand in this moment
if you are unfortunate enough to have SLMs in your life, here is my best attempt at compassionately reading them for filth
medium.com/prismnpen/lo...
if you are unfortunate enough to have SLMs in your life, here is my best attempt at compassionately reading them for filth
medium.com/prismnpen/lo...
Got a new Mac! M5 Max with 64GB RAM
Anyone using local SLMs with coding agents on this hardware? Wondering how far I can go with local models.
Anyone using local SLMs with coding agents on this hardware? Wondering how far I can go with local models.
May 7, 2026 at 9:50 PM
Got a new Mac! M5 Max with 64GB RAM
Anyone using local SLMs with coding agents on this hardware? Wondering how far I can go with local models.
Anyone using local SLMs with coding agents on this hardware? Wondering how far I can go with local models.
Seems like a good potential compromise who find these things useful. Also note: SLMs do not babysit people at all or pretend to do “personal care.”Pretty big case of feature not a bug.
September 11, 2026 at 9:19 PM
Seems like a good potential compromise who find these things useful. Also note: SLMs do not babysit people at all or pretend to do “personal care.”Pretty big case of feature not a bug.
Awesome SLMs: a List of small LLMs that can run on Raspberry Pi5 (8GB RAM) with Ollama - Project by Philippe Charrière @k33gorg.bsky.social #LLM #AI
Awesome SLMs
This is the list of the SLMs I use on my Raspberry Pi5 (8GB RAM) with Ollama
parakeet-nest.github.io
February 21, 2025 at 8:05 AM
Awesome SLMs: a List of small LLMs that can run on Raspberry Pi5 (8GB RAM) with Ollama - Project by Philippe Charrière @k33gorg.bsky.social #LLM #AI
Uwu-izing/furry-izing macro and string replacement systems are an example of SLMs (smol language models)
May 24, 2024 at 6:06 PM
Uwu-izing/furry-izing macro and string replacement systems are an example of SLMs (smol language models)
They will try to get you alone at all costs, at this point.
That is because they are SLMs who fear public disapproval. That is their weakness, esp when they try to spar with anyone who has been systemically marginalized, and so hardened to public disapproval.
That is because they are SLMs who fear public disapproval. That is their weakness, esp when they try to spar with anyone who has been systemically marginalized, and so hardened to public disapproval.
January 25, 2025 at 2:05 AM
They will try to get you alone at all costs, at this point.
That is because they are SLMs who fear public disapproval. That is their weakness, esp when they try to spar with anyone who has been systemically marginalized, and so hardened to public disapproval.
That is because they are SLMs who fear public disapproval. That is their weakness, esp when they try to spar with anyone who has been systemically marginalized, and so hardened to public disapproval.