Ankur Guoswami
@ankurgos.bsky.social
Living in and probably posting about seattle. Warriors fan, ml engineer (posts my own yada yada)
Reposted by Ankur Guoswami
BREAKING Greg Bovino has been removed as Border Patrol "commander at large" and will return to El Centro Calif, where he is expected to retire soon. A stunning turnaround after Pretti killing. Bovino's traveling blue city crackdown is over www.theatlantic.com/politics/202...
Gregory Bovino Gets Demoted
The Border Patrol chief was the public face of a traveling immigration crackdown on cities governed by Democrats.
www.theatlantic.com
January 26, 2026 at 11:40 PM
BREAKING Greg Bovino has been removed as Border Patrol "commander at large" and will return to El Centro Calif, where he is expected to retire soon. A stunning turnaround after Pretti killing. Bovino's traveling blue city crackdown is over www.theatlantic.com/politics/202...
feeling good go hawks
January 26, 2026 at 4:06 AM
feeling good go hawks
Reposted by Ankur Guoswami
an insightful account/parable on this, well worth a read : www.galois.com/articles/spe...
January 23, 2026 at 4:57 PM
an insightful account/parable on this, well worth a read : www.galois.com/articles/spe...
Reposted by Ankur Guoswami
"serious doubts over >50% of high impact papers [on] microplastics...brain microplastic paper is a joke...race to publish led to rushed results and routine checks overlooked" www.theguardian.com/environment/...
‘A bombshell’: doubt cast on discovery of microplastics throughout human body
Exclusive: Some scientists say many detections are most likely error, with one high-profile study called a ‘joke’
www.theguardian.com
January 13, 2026 at 5:12 PM
"serious doubts over >50% of high impact papers [on] microplastics...brain microplastic paper is a joke...race to publish led to rushed results and routine checks overlooked" www.theguardian.com/environment/...
Reposted by Ankur Guoswami
Introducing DroPE: Extending Context by Dropping Positional Embeddings
We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity.
arxiv.org/abs/2512.12167
pub.sakana.ai/DroPE
We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity.
arxiv.org/abs/2512.12167
pub.sakana.ai/DroPE
January 12, 2026 at 4:07 AM
Introducing DroPE: Extending Context by Dropping Positional Embeddings
We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity.
arxiv.org/abs/2512.12167
pub.sakana.ai/DroPE
We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity.
arxiv.org/abs/2512.12167
pub.sakana.ai/DroPE
Reposted by Ankur Guoswami
Here's your new mayor, Seattle.
January 2, 2026 at 7:00 PM
Here's your new mayor, Seattle.
Reposted by Ankur Guoswami
Jeremy’s post got me thinking about a car-free Fremont Bridge. If @seattledot.bsky.social and @wilsonforseattle.bsky.social are serious about mode shift, this would be a great project. seattletransitblog.com/2025/12/22/t...
We should build for the city we want, not the city we have.
We should build for the city we want, not the city we have.
December 22, 2025 at 9:22 PM
Jeremy’s post got me thinking about a car-free Fremont Bridge. If @seattledot.bsky.social and @wilsonforseattle.bsky.social are serious about mode shift, this would be a great project. seattletransitblog.com/2025/12/22/t...
We should build for the city we want, not the city we have.
We should build for the city we want, not the city we have.
Reposted by Ankur Guoswami
To teach us how modern LLM inference works, LMSYS gave us a readable version of SGLang where they distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance.
mini-sglang
github.com/sgl-project/...
mini-sglang
github.com/sgl-project/...
December 19, 2025 at 7:40 AM
To teach us how modern LLM inference works, LMSYS gave us a readable version of SGLang where they distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance.
mini-sglang
github.com/sgl-project/...
mini-sglang
github.com/sgl-project/...
Reposted by Ankur Guoswami
Wrapping up reinforcement learning week on the blog: Why RL is always leaving efficiency on the table, and what you can do about it.
There's got to be a better way!
From Reformist RL to the principle of certainty equivalence.
www.argmin.net
December 5, 2025 at 3:29 PM
Wrapping up reinforcement learning week on the blog: Why RL is always leaving efficiency on the table, and what you can do about it.
Reposted by Ankur Guoswami
We present Olmo 3, our next family of fully open, leading language models.
This family of 7B and 32B models represents:
1. The best 32B base model.
2. The best 7B Western thinking & instruct models.
3. The first 32B (or larger) fully open reasoning model.
This family of 7B and 32B models represents:
1. The best 32B base model.
2. The best 7B Western thinking & instruct models.
3. The first 32B (or larger) fully open reasoning model.
November 20, 2025 at 2:32 PM
We present Olmo 3, our next family of fully open, leading language models.
This family of 7B and 32B models represents:
1. The best 32B base model.
2. The best 7B Western thinking & instruct models.
3. The first 32B (or larger) fully open reasoning model.
This family of 7B and 32B models represents:
1. The best 32B base model.
2. The best 7B Western thinking & instruct models.
3. The first 32B (or larger) fully open reasoning model.
@wilsonforseattle.bsky.social wins!!!
November 11, 2025 at 1:25 AM
@wilsonforseattle.bsky.social wins!!!
Pluribus is so good. Great first couple episodes
November 10, 2025 at 7:49 AM
Pluribus is so good. Great first couple episodes
the ad campaign coming in for harrell this last week is crazy. Honestly a gross campaign
November 3, 2025 at 12:33 AM
the ad campaign coming in for harrell this last week is crazy. Honestly a gross campaign
Reposted by Ankur Guoswami
I guess we’re at the point in the campaign where pundits and consultants make excuses as to why their candidate is losing. 🧵
Westneat points to Trump and his fawning billionaires:
Westneat points to Trump and his fawning billionaires:
The switch that flipped in Seattle politics
The sight of unimaginable Big Tech wealth at the gates of American democracy shook Seattle politics. The mayor, among others, may not survive the quake.
www.seattletimes.com
November 1, 2025 at 6:06 PM
I guess we’re at the point in the campaign where pundits and consultants make excuses as to why their candidate is losing. 🧵
Westneat points to Trump and his fawning billionaires:
Westneat points to Trump and his fawning billionaires:
Reposted by Ankur Guoswami
Harrell Argues Seattle Needs Mayor Who Can Afford Enough Childcare to Forget Own Child Even Exists: tinyurl.com/2sjh2ca5
October 31, 2025 at 7:16 PM
Harrell Argues Seattle Needs Mayor Who Can Afford Enough Childcare to Forget Own Child Even Exists: tinyurl.com/2sjh2ca5
Reposted by Ankur Guoswami
Seattle No Kings from the monorail
October 18, 2025 at 10:20 PM
Seattle No Kings from the monorail
Reposted by Ankur Guoswami
Here is a story about a AAAI paper that got 5 human reviews and 1 AI review. The AI generated review contained a page-long counterexample to a proof in the paper. The AI’s counter example contained hallucinated errors. How does one rebut an AI reviewer? www.linkedin.com/posts/omerbp...
#aaai26 | Omer Ben-Porat | 13 comments
#AAAI26
This year, in addition to five human reviewers, we also received an AI review.
The humans sort of liked the paper, so there's hope.
The AI, however, took a bolder approach: it confidently d...
www.linkedin.com
October 16, 2025 at 1:19 AM
Here is a story about a AAAI paper that got 5 human reviews and 1 AI review. The AI generated review contained a page-long counterexample to a proof in the paper. The AI’s counter example contained hallucinated errors. How does one rebut an AI reviewer? www.linkedin.com/posts/omerbp...
Reposted by Ankur Guoswami
Inspired to share some papers that I found at #COLM2025!
"Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation" by Amanda Myntti et al. arxiv.org/abs/2504.01542
"Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation" by Amanda Myntti et al. arxiv.org/abs/2504.01542
October 14, 2025 at 6:16 PM
Inspired to share some papers that I found at #COLM2025!
"Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation" by Amanda Myntti et al. arxiv.org/abs/2504.01542
"Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation" by Amanda Myntti et al. arxiv.org/abs/2504.01542
Reposted by Ankur Guoswami
Reposted by Ankur Guoswami
🚨New Preprint: We develop a novel task that probes counterfactual thinking without using counterfactual language, and that teases apart genuine counterfactual thinking from related forms of thinking. Using this task, we find that the ability for counterfactual thinking emerges around 5 years of age.
October 13, 2025 at 7:58 PM
🚨New Preprint: We develop a novel task that probes counterfactual thinking without using counterfactual language, and that teases apart genuine counterfactual thinking from related forms of thinking. Using this task, we find that the ability for counterfactual thinking emerges around 5 years of age.
Reposted by Ankur Guoswami
By popular demand, @aceckhouse.bsky.social and I have put together a deep-in-the-weeds explainer on SB 79. Enjoy!
substack.com/home/post/p-...
substack.com/home/post/p-...
Everything You Need To Know About SB 79
…but were too afraid to read about in the actual bill.
substack.com
October 12, 2025 at 7:37 PM
By popular demand, @aceckhouse.bsky.social and I have put together a deep-in-the-weeds explainer on SB 79. Enjoy!
substack.com/home/post/p-...
substack.com/home/post/p-...
MARINERS BABYYYYY
October 11, 2025 at 5:09 AM
MARINERS BABYYYYY
Reposted by Ankur Guoswami
💡Can we trust synthetic data for statistical inference?
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
October 10, 2025 at 4:12 PM
💡Can we trust synthetic data for statistical inference?
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
We show that synthetic data (e.g., LLM simulations) can significantly improve the performance of inference tasks. The key intuition lies in the interactions between the moment residuals of synthetic data and those of real data
Reposted by Ankur Guoswami
MAJOR NEWS: @GavinNewsom signed SB 79, my bill allowing more housing near public transit — rail, subway, rapid bus.
It’s a huge step for housing in California. It’ll create more homes, strengthen our transit systems & reduce traffic & carbon emissions.
Thank you, Governor!
It’s a huge step for housing in California. It’ll create more homes, strengthen our transit systems & reduce traffic & carbon emissions.
Thank you, Governor!
October 10, 2025 at 4:37 PM
MAJOR NEWS: @GavinNewsom signed SB 79, my bill allowing more housing near public transit — rail, subway, rapid bus.
It’s a huge step for housing in California. It’ll create more homes, strengthen our transit systems & reduce traffic & carbon emissions.
Thank you, Governor!
It’s a huge step for housing in California. It’ll create more homes, strengthen our transit systems & reduce traffic & carbon emissions.
Thank you, Governor!