Ismael Velasco
@ismael-velasco.bsky.social
Involuntary polymath.
Writer, poet, storyteller, mime.
Software engineer (green compute, Web & AI).
Peer reviewed author (sustainability, humanities, social sciences).
Social activist.
Inadequate human & Bahá'í.
linkedin.com/in/IsmaelVelasco
Writer, poet, storyteller, mime.
Software engineer (green compute, Web & AI).
Peer reviewed author (sustainability, humanities, social sciences).
Social activist.
Inadequate human & Bahá'í.
linkedin.com/in/IsmaelVelasco
Pinned
Hope is the boundary of action. The engine or the limit of realisation. If I cannot hope to leave my house, I will plan to live my whole life in its rooms.
Hope is not selective blindness, but a prerequisite of sight. It offers no guarantees, only motion. No solution is conceivable without it.
Hope is not selective blindness, but a prerequisite of sight. It offers no guarantees, only motion. No solution is conceivable without it.
Reposted by Ismael Velasco
8/8 Anyways, it's nice paper and a refreshing read in the era of LLMs: arxiv.org/abs/2510.04871
Less is More: Recursive Reasoning with Tiny Networks
Hierarchical Reasoning Model (HRM) is a novel approach using two small neural networks recursing at different frequencies. This biologically inspired method beats Large Language models (LLMs) on hard ...
arxiv.org
October 9, 2025 at 4:23 PM
8/8 Anyways, it's nice paper and a refreshing read in the era of LLMs: arxiv.org/abs/2510.04871
Reposted by Ismael Velasco
From the Hierarchical Reasoning Model (HRM) to a new Tiny Recursive Model (TRM).
A few months ago, the HRM made big waves in the AI research community as it showed really good performance on the ARC challenge despite its small 27M size. (That's about 22x smaller than the smallest Qwen3 0.6B model.)
A few months ago, the HRM made big waves in the AI research community as it showed really good performance on the ARC challenge despite its small 27M size. (That's about 22x smaller than the smallest Qwen3 0.6B model.)
October 9, 2025 at 4:23 PM
From the Hierarchical Reasoning Model (HRM) to a new Tiny Recursive Model (TRM).
A few months ago, the HRM made big waves in the AI research community as it showed really good performance on the ARC challenge despite its small 27M size. (That's about 22x smaller than the smallest Qwen3 0.6B model.)
A few months ago, the HRM made big waves in the AI research community as it showed really good performance on the ARC challenge despite its small 27M size. (That's about 22x smaller than the smallest Qwen3 0.6B model.)
Reposted by Ismael Velasco
Updated & turned my Big LLM Architecture Comparison article into a video lecture.
The 11 LLM archs covered in this video:
1. DeepSeek V3/R1
2. OLMo 2
3. Gemma 3
4. Mistral Small 3.1
5. Llama 4
6. Qwen3
7. SmolLM3
8. Kimi 2
9. GPT-OSS
10. Grok 2.5
11. GLM-4.5/4.6
www.youtube.com/watch?v=rNlU...
The 11 LLM archs covered in this video:
1. DeepSeek V3/R1
2. OLMo 2
3. Gemma 3
4. Mistral Small 3.1
5. Llama 4
6. Qwen3
7. SmolLM3
8. Kimi 2
9. GPT-OSS
10. Grok 2.5
11. GLM-4.5/4.6
www.youtube.com/watch?v=rNlU...
The Big LLM Architecture Comparison
YouTube video by Sebastian Raschka
www.youtube.com
October 10, 2025 at 5:05 PM
Updated & turned my Big LLM Architecture Comparison article into a video lecture.
The 11 LLM archs covered in this video:
1. DeepSeek V3/R1
2. OLMo 2
3. Gemma 3
4. Mistral Small 3.1
5. Llama 4
6. Qwen3
7. SmolLM3
8. Kimi 2
9. GPT-OSS
10. Grok 2.5
11. GLM-4.5/4.6
www.youtube.com/watch?v=rNlU...
The 11 LLM archs covered in this video:
1. DeepSeek V3/R1
2. OLMo 2
3. Gemma 3
4. Mistral Small 3.1
5. Llama 4
6. Qwen3
7. SmolLM3
8. Kimi 2
9. GPT-OSS
10. Grok 2.5
11. GLM-4.5/4.6
www.youtube.com/watch?v=rNlU...
Reposted by Ismael Velasco
A short talk on the main architecture components of LLMs this year + a look beyond the transformer architecture: www.youtube.com/watch?v=lONy...
October 27, 2025 at 3:45 PM
A short talk on the main architecture components of LLMs this year + a look beyond the transformer architecture: www.youtube.com/watch?v=lONy...
Reposted by Ismael Velasco
I just saw the Kimi K2 Thinking release!
Kimi K2 is based on the DeepSeek V3/R1 architecture, and here's a side-by-side comparison.
In short, Kimi K2 is a slightly scaled DeepSeek V3/R1. And the gains are in the data and training recipes. Hopefully, we will see some details on those soon, too.
Kimi K2 is based on the DeepSeek V3/R1 architecture, and here's a side-by-side comparison.
In short, Kimi K2 is a slightly scaled DeepSeek V3/R1. And the gains are in the data and training recipes. Hopefully, we will see some details on those soon, too.
November 6, 2025 at 7:35 PM
I just saw the Kimi K2 Thinking release!
Kimi K2 is based on the DeepSeek V3/R1 architecture, and here's a side-by-side comparison.
In short, Kimi K2 is a slightly scaled DeepSeek V3/R1. And the gains are in the data and training recipes. Hopefully, we will see some details on those soon, too.
Kimi K2 is based on the DeepSeek V3/R1 architecture, and here's a side-by-side comparison.
In short, Kimi K2 is a slightly scaled DeepSeek V3/R1. And the gains are in the data and training recipes. Hopefully, we will see some details on those soon, too.
Reposted by Ismael Velasco
My new field guide to alternatives to standard LLMs:
Gated DeltaNet hybrids (Qwen3-Next, Kimi Linear), text diffusion, code world models, and small reasoning transformers.
🔗 magazine.sebastianraschka.com/p/beyond-sta...
Gated DeltaNet hybrids (Qwen3-Next, Kimi Linear), text diffusion, code world models, and small reasoning transformers.
🔗 magazine.sebastianraschka.com/p/beyond-sta...
November 4, 2025 at 2:49 PM
My new field guide to alternatives to standard LLMs:
Gated DeltaNet hybrids (Qwen3-Next, Kimi Linear), text diffusion, code world models, and small reasoning transformers.
🔗 magazine.sebastianraschka.com/p/beyond-sta...
Gated DeltaNet hybrids (Qwen3-Next, Kimi Linear), text diffusion, code world models, and small reasoning transformers.
🔗 magazine.sebastianraschka.com/p/beyond-sta...
Reposted by Ismael Velasco
Poetry as a vector for breaking AI constraints—verse prompts jailbreak LLMs up to 18x more effectively than prose (90%+ in some models). A fascinating case study in how aesthetic structures interact with computational control systems.
arxiv.org/abs/2511.15304
arxiv.org/abs/2511.15304
Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models
We present evidence that adversarial poetry functions as a universal single-turn jailbreak technique for Large Language Models (LLMs). Across 25 frontier proprietary and open-weight models, curated po...
arxiv.org
November 23, 2025 at 4:47 AM
Poetry as a vector for breaking AI constraints—verse prompts jailbreak LLMs up to 18x more effectively than prose (90%+ in some models). A fascinating case study in how aesthetic structures interact with computational control systems.
arxiv.org/abs/2511.15304
arxiv.org/abs/2511.15304
Reposted by Ismael Velasco
I think Claude Code has achieved AGI
November 22, 2025 at 9:09 PM
I think Claude Code has achieved AGI
Reposted by Ismael Velasco
The Latent Role of Open Models in the AI Economy - papers.ssrn.com/sol3/papers.... move to open would "generate an estimated $24.8 billion in additional consumer savings across 2025" (v @azeem.bsky.social) #openness
The Latent Role of Open Models in the AI Economy
The rapid diffusion of large language models (LLMs) is mediated by an emerging market for AI inference. However, its economic structure is poorly understood due
papers.ssrn.com
November 23, 2025 at 10:39 AM
The Latent Role of Open Models in the AI Economy - papers.ssrn.com/sol3/papers.... move to open would "generate an estimated $24.8 billion in additional consumer savings across 2025" (v @azeem.bsky.social) #openness
Reposted by Ismael Velasco
⚡ Hackernews Top story: An Economy of AI Agents
An Economy of AI Agents
In the coming decade, artificially intelligent agents with the ability to plan and execute complex tasks over long time horizons with little direct oversight from humans may be deployed across the economy. This chapter surveys recent developments and highlights open questions for economists around how AI agents might interact with humans and with each other, shape markets and organizations, and what institutions might be required for well-functioning markets.
arxiv.org
November 23, 2025 at 7:20 AM
⚡ Hackernews Top story: An Economy of AI Agents
Reposted by Ismael Velasco
Made a page on my website with instructions on how to disable AI features for a variety of services (mainly the ones I encounter at the moment, will append to the list as I go).
The list so far includes Github, Google Services, Firefox and DuckDuckGo.
kokorobot.ca/site/remove_...
The list so far includes Github, Google Services, Firefox and DuckDuckGo.
kokorobot.ca/site/remove_...
Rek Bell — remove ai
kokorobot.ca
November 22, 2025 at 6:30 PM
Made a page on my website with instructions on how to disable AI features for a variety of services (mainly the ones I encounter at the moment, will append to the list as I go).
The list so far includes Github, Google Services, Firefox and DuckDuckGo.
kokorobot.ca/site/remove_...
The list so far includes Github, Google Services, Firefox and DuckDuckGo.
kokorobot.ca/site/remove_...
Sycophantic AI agrees 50% more than humans even with harmful conversations: arxiv.org/abs/2510.01395
October 11, 2025 at 9:46 AM
Sycophantic AI agrees 50% more than humans even with harmful conversations: arxiv.org/abs/2510.01395
Reposted by Ismael Velasco
NYC Takes on Big Tech in Youth Mental Health Fight
NYC Takes on Big Tech in Youth Mental Health Fight
New York City has filed a sweeping lawsuit against some of the world's biggest social media companies, accusing them of driving a youth mental health crisis that has strained schools and public hospitals, The Hill reported.
dlvr.it
October 9, 2025 at 11:55 PM
NYC Takes on Big Tech in Youth Mental Health Fight
AI for scientific discovery is a social problem, and just adding more compute per Sam Altman may not actually cure cancer.
arxiv.org/html/2509.06...
arxiv.org/html/2509.06...
October 8, 2025 at 7:14 AM
AI for scientific discovery is a social problem, and just adding more compute per Sam Altman may not actually cure cancer.
arxiv.org/html/2509.06...
arxiv.org/html/2509.06...
Reposted by Ismael Velasco
$380 million - the amount the Senate wants to allocate just for AI and automation for the department of defense
Do you think they'll earn that back, or is that just a sunk cost for us to pay the government to track our data?
Do you think they'll earn that back, or is that just a sunk cost for us to pay the government to track our data?
July 2, 2025 at 3:30 PM
$380 million - the amount the Senate wants to allocate just for AI and automation for the department of defense
Do you think they'll earn that back, or is that just a sunk cost for us to pay the government to track our data?
Do you think they'll earn that back, or is that just a sunk cost for us to pay the government to track our data?
Reposted by Ismael Velasco
AI and automation could eliminate nearly 100 million jobs in the U.S. in the next decade, a report set to be released by the Senate on Monday finds. #climatechange #climatecrisis #globalwarming
www.axios.com/2025/10/06/a...
www.axios.com/2025/10/06/a...
Exclusive: AI could erase 100 million U.S. jobs, Senate Dem report finds
A ChatGPT-based analysis from Democrats on the Senate Health, Education, Labor and Pensions Committee found that AI could wipe out swaths of both white- and blue-collar jobs
www.axios.com
October 6, 2025 at 3:29 PM
AI and automation could eliminate nearly 100 million jobs in the U.S. in the next decade, a report set to be released by the Senate on Monday finds. #climatechange #climatecrisis #globalwarming
www.axios.com/2025/10/06/a...
www.axios.com/2025/10/06/a...
I enjoyed this nuanced, humble and thought provoking paper on what I would call (surely not the first) AI displacement anxiety. This focuses on philosophy, but broadens the frame toward the end. I think I might start a list on displacement anxiety, pieces like this feel like canaries in the mine
In which I throw a small and very timid cat amongst some large and aggressive pigeons.
philpapers.org/rec/LEVTWO-10
philpapers.org/rec/LEVTWO-10
Neil Levy, The work of philosophy in the age of mechanical reproduction: How AI threatens the meaningfulness of philosophy - PhilPapers
Some philosophers have worried that AI threatens meaning in life by making humans redundant in the workplace or by relegating them to menial tasks. While these worries may be justified, there's ...
philpapers.org
October 6, 2025 at 11:58 AM
I enjoyed this nuanced, humble and thought provoking paper on what I would call (surely not the first) AI displacement anxiety. This focuses on philosophy, but broadens the frame toward the end. I think I might start a list on displacement anxiety, pieces like this feel like canaries in the mine
Reposted by Ismael Velasco
Challenge Poverty Week
Day 1: Social Security - Building a Foundation for a Better Life
As part of Challenge Poverty Week, we are drawing attention to the vital financial support available to families through Social Security Scotland.
Visit the website for more info:
www.socialsecurity.gov.scot
Day 1: Social Security - Building a Foundation for a Better Life
As part of Challenge Poverty Week, we are drawing attention to the vital financial support available to families through Social Security Scotland.
Visit the website for more info:
www.socialsecurity.gov.scot
October 6, 2025 at 10:15 AM
Challenge Poverty Week
Day 1: Social Security - Building a Foundation for a Better Life
As part of Challenge Poverty Week, we are drawing attention to the vital financial support available to families through Social Security Scotland.
Visit the website for more info:
www.socialsecurity.gov.scot
Day 1: Social Security - Building a Foundation for a Better Life
As part of Challenge Poverty Week, we are drawing attention to the vital financial support available to families through Social Security Scotland.
Visit the website for more info:
www.socialsecurity.gov.scot
The headline is click-bait, but the second half of the article summarises well some core debates on intelligence in relation to, and contrast with, LLMs. Several useful concepts and voices to explore further.
www.forbes.com/sites/hessie...
www.forbes.com/sites/hessie...
LLMs Are Short-Circuiting. Is It Time To Redefine Intelligence?
The AI community is going through a reckoning. LLMs are failing. They are proving they may not be the pathway towards the promised intelligence we've all been led to believe. Now what?
www.forbes.com
October 6, 2025 at 1:18 AM
The headline is click-bait, but the second half of the article summarises well some core debates on intelligence in relation to, and contrast with, LLMs. Several useful concepts and voices to explore further.
www.forbes.com/sites/hessie...
www.forbes.com/sites/hessie...
Reposted by Ismael Velasco
It's on its way and over 100 people are already booked on. Are you? More here: dgcommunity.tsdg.org.uk
October 3, 2025 at 2:52 PM
It's on its way and over 100 people are already booked on. Are you? More here: dgcommunity.tsdg.org.uk
Reposted by Ismael Velasco
September 29, 2025 at 11:07 AM
An excellent, innovative, revisionist study highlighting continuities in work patterns pre/post factory era. Challenges received wisdom while visibilising the labour of women. I wish it explored discontinuities with industrial/industrious revolution that remain: what survives from de Vries/Thompson?
Who did what in early modern England?
New #OpenAccess book, 'The Experience of Work in Early Modern England' by @jwhittle.bsky.social, @markhailwood.bsky.social, @hkrobb.bsky.social & @aucointaylor.bsky.social, based on thousands of #EarlyModern court depositions 🗃️
Read it: doi.org/10.1017/9781...
New #OpenAccess book, 'The Experience of Work in Early Modern England' by @jwhittle.bsky.social, @markhailwood.bsky.social, @hkrobb.bsky.social & @aucointaylor.bsky.social, based on thousands of #EarlyModern court depositions 🗃️
Read it: doi.org/10.1017/9781...
October 2, 2025 at 10:27 AM
An excellent, innovative, revisionist study highlighting continuities in work patterns pre/post factory era. Challenges received wisdom while visibilising the labour of women. I wish it explored discontinuities with industrial/industrious revolution that remain: what survives from de Vries/Thompson?
This is a very important point on the impact of adopting the language of the labs when describing the risks and problematic impacts of AI.
www.linkedin.com/posts/ismael...
www.linkedin.com/posts/ismael...
September 27, 2025 at 12:36 AM
This is a very important point on the impact of adopting the language of the labs when describing the risks and problematic impacts of AI.
www.linkedin.com/posts/ismael...
www.linkedin.com/posts/ismael...
I enjoyed, and resonated, with this perspective on the parameters, possibilities and constraints of putative LLM consciousness. An unaddressed area for me is anything like selfhood or coherent/persistent identity. Without exogenous systems the llm model alone is ab nihil each prompt.
Here's a post copying over my summary thoughts from a consciousness science group text where folks were trying to build some consensus (this is all me—there was not a whole lot of consensus reached, because there's such a diversity of opinions): leaflet.pub/dc2d21b7-6c1...
September 23, 2025 at 11:57 PM
I enjoyed, and resonated, with this perspective on the parameters, possibilities and constraints of putative LLM consciousness. An unaddressed area for me is anything like selfhood or coherent/persistent identity. Without exogenous systems the llm model alone is ab nihil each prompt.