Peter
@petmouse.bluesky.mousses.xyz
I like computers and nature 🐨☦️
Pinned
Reposted by Peter
‘The biggest news regarding the release of Gemini 3 was buried in the methods: Google got better results than it is competitors without using Nvidia GPUs, relying solely on their own TPUs.‘ garymarcus.substack.com/p/hot-take-o... by @garymarcus.bsky.social
November 19, 2025 at 5:16 PM
‘The biggest news regarding the release of Gemini 3 was buried in the methods: Google got better results than it is competitors without using Nvidia GPUs, relying solely on their own TPUs.‘ garymarcus.substack.com/p/hot-take-o... by @garymarcus.bsky.social
Reposted by Peter
investments in solar power lower costs for consumers so much that in some places they're literally giving away free energy during the day, but sure.
it won’t feel good to say but it is crucial to actually admit this:
democrats’ pivot to affordability is already seeing them tear up climate policies
and any effort to lower costs for consumers will conflict with investments in renewables, unless even more money is poured into them
democrats’ pivot to affordability is already seeing them tear up climate policies
and any effort to lower costs for consumers will conflict with investments in renewables, unless even more money is poured into them
Democrats’ New Affordability Message Is Complicating Climate Policy
The transition to clean energy will be expensive today, even if it’ll be cheaper in the long run.
heatmap.news
November 19, 2025 at 2:50 PM
investments in solar power lower costs for consumers so much that in some places they're literally giving away free energy during the day, but sure.
Reposted by Peter
Reposted by Peter
On November 18 Cloudflare experienced a service outage. Here's a detailed breakdown of what happened. https://cfl.re/43Bw8AI
Cloudflare outage on November 18, 2025
Cloudflare suffered a service outage on November 18, 2025. The outage was triggered by a bug in generation logic for a Bot Management feature file causing many Cloudflare services to be affected.
blog.cloudflare.com
November 18, 2025 at 11:49 PM
On November 18 Cloudflare experienced a service outage. Here's a detailed breakdown of what happened. https://cfl.re/43Bw8AI
Reposted by Peter
software engineering has taught me that it's a lot easier to withstand a nuclear attack than a fat-fingered config
The internet was designed to withstand a nuclear attack, and now has difficulty withstanding a fat fingered config.
November 19, 2025 at 12:06 AM
software engineering has taught me that it's a lot easier to withstand a nuclear attack than a fat-fingered config
Reposted by Peter
Gemini 3 DeepThink is 2x SOTA on ARC-AGI-2
November 18, 2025 at 8:59 PM
Gemini 3 DeepThink is 2x SOTA on ARC-AGI-2
Reposted by Peter
Gemini 3 is indeed a much larger model
both pre-training (model size) & post-training (RL) scaling
both pre-training (model size) & post-training (RL) scaling
November 18, 2025 at 9:21 PM
Gemini 3 is indeed a much larger model
both pre-training (model size) & post-training (RL) scaling
both pre-training (model size) & post-training (RL) scaling
Reposted by Peter
Reposted by Peter
New: DHS claimed in court proceedings that nearly two weeks worth of surveillance footage from ICE’s Broadview Detention Center in suburban Chicago has been “irretrievably destroyed” and may not be able to be recovered. ACLU of Illinois said "Hoping DHS can explain"
www.404media.co/two-weeks-of...
www.404media.co/two-weeks-of...
Two Weeks of Surveillance Footage From ICE Detention Center ‘Irretrievably Destroyed’
"Defendants have indicated that some video between October 19, 2025 and October 31, 2025 has been irretrievably destroyed and therefore cannot be produced on an expedited basis or at all."
www.404media.co
November 18, 2025 at 5:03 PM
New: DHS claimed in court proceedings that nearly two weeks worth of surveillance footage from ICE’s Broadview Detention Center in suburban Chicago has been “irretrievably destroyed” and may not be able to be recovered. ACLU of Illinois said "Hoping DHS can explain"
www.404media.co/two-weeks-of...
www.404media.co/two-weeks-of...
Reposted by Peter
Jeff selling shirts might be the new Pizza Index but for internet problems
November 18, 2025 at 1:52 PM
Jeff selling shirts might be the new Pizza Index but for internet problems
Reposted by Peter
NEW: Internet infrastructure giant Cloudflare blamed this morning's massive internet outage on a "latent bug."
This is another stark reminder that the internet depends on just a handful of companies. According to an estimate, Cloudflare is used by 20% of all websites on the internet.
This is another stark reminder that the internet depends on just a handful of companies. According to an estimate, Cloudflare is used by 20% of all websites on the internet.
Cloudflare blames massive internet outage on 'latent bug' | TechCrunch
An outage at internet infrastructure giant Cloudflare took down several big websites and services, including ChatGPT, Claude, Spotify, and X.
techcrunch.com
November 18, 2025 at 3:44 PM
NEW: Internet infrastructure giant Cloudflare blamed this morning's massive internet outage on a "latent bug."
This is another stark reminder that the internet depends on just a handful of companies. According to an estimate, Cloudflare is used by 20% of all websites on the internet.
This is another stark reminder that the internet depends on just a handful of companies. According to an estimate, Cloudflare is used by 20% of all websites on the internet.
Reposted by Peter
Announcing RefCOCO-M, a refreshed RefCOCO with pixel-accurate masks and the problematic prompts removed. Better data for better evaluation.
huggingface.co/datasets/moo...
huggingface.co/datasets/moo...
November 18, 2025 at 6:55 AM
Announcing RefCOCO-M, a refreshed RefCOCO with pixel-accurate masks and the problematic prompts removed. Better data for better evaluation.
huggingface.co/datasets/moo...
huggingface.co/datasets/moo...
Reposted by Peter
lol he sounds like the fucking Penguin here
"Quiet! Quiet, Piggy."
November 18, 2025 at 3:16 AM
lol he sounds like the fucking Penguin here
Reposted by Peter
Oh thanks for the reminder. We released the learning SDK to add state to ~whatever, looking for feedback:
forum.letta.com/t/announcing...
forum.letta.com/t/announcing...
Announcing the Agentic Learning SDK: Add state to anything
This is huge for adoption - it removes the biggest friction point for teams wanting Letta’s memory capabilities without refactoring existing applications. Key Insight: The context manager pattern (w...
forum.letta.com
November 16, 2025 at 7:24 PM
Oh thanks for the reminder. We released the learning SDK to add state to ~whatever, looking for feedback:
forum.letta.com/t/announcing...
forum.letta.com/t/announcing...
Reposted by Peter
So glad BSky lets me declare my interests so they can show me more of what I li….
November 17, 2025 at 12:39 PM
So glad BSky lets me declare my interests so they can show me more of what I li….
Reposted by Peter
Our port to Android 16 QPR1 which was to AOSP on November 11 is currently being tested internally. Several important regressions have been discovered and we're working on resolving those before we release it for public testing. A few minor features also still need to be ported.
November 16, 2025 at 8:35 PM
Our port to Android 16 QPR1 which was to AOSP on November 11 is currently being tested internally. Several important regressions have been discovered and we're working on resolving those before we release it for public testing. A few minor features also still need to be ported.
Reposted by Peter
From a CTO at a startup:
"We interview devs by giving them a task to build an app on the spot, from scratch (2x BE endpoints, some frontend.) They can use AI, ofc - and we dig into why they did this or that.
What is surprising: 14/15 devs from Meta failed this screening."
"We interview devs by giving them a task to build an app on the spot, from scratch (2x BE endpoints, some frontend.) They can use AI, ofc - and we dig into why they did this or that.
What is surprising: 14/15 devs from Meta failed this screening."
November 14, 2025 at 10:53 PM
From a CTO at a startup:
"We interview devs by giving them a task to build an app on the spot, from scratch (2x BE endpoints, some frontend.) They can use AI, ofc - and we dig into why they did this or that.
What is surprising: 14/15 devs from Meta failed this screening."
"We interview devs by giving them a task to build an app on the spot, from scratch (2x BE endpoints, some frontend.) They can use AI, ofc - and we dig into why they did this or that.
What is surprising: 14/15 devs from Meta failed this screening."
Reposted by Peter
i feel like America is very behind in every way that matters
Robots have been getting really dynamic and capable this year while no one was looking
November 16, 2025 at 3:24 PM
i feel like America is very behind in every way that matters
Hot take: we already have AGI, its just that people overestimate general intelligence
Let's recognize that peoples timelines for AGI have moved back. Those that held the shortest timelines should say "I was a little wrong" and ask why they were wrong.
> the authors of AI 2027 think that their “strong AGI in 2027” scenario is plausible, but faster than their median expectations. The median timelines of Daniel, Eli, and Thomas for when we will develop strong AGI are 2030, 2035, and 2033, respectively.
November 16, 2025 at 5:33 PM
Hot take: we already have AGI, its just that people overestimate general intelligence
Reposted by Peter
Physics of Language Models: Part 3.1
If you show a fact to an LLM in pre-training once, it’ll memorize the form but not the fact itself.
but if you (synthetically) rephrase the text several times, it’ll memorize the fact
arxiv.org/abs/2309.14316
If you show a fact to an LLM in pre-training once, it’ll memorize the form but not the fact itself.
but if you (synthetically) rephrase the text several times, it’ll memorize the fact
arxiv.org/abs/2309.14316
Physics of Language Models: Part 3.1, Knowledge Storage and Extraction
Large language models (LLMs) can store a vast amount of world knowledge, often extractable via question-answering (e.g., "What is Abraham Lincoln's birthday?"). However, do they answer such questions ...
arxiv.org
November 16, 2025 at 3:01 PM
Physics of Language Models: Part 3.1
If you show a fact to an LLM in pre-training once, it’ll memorize the form but not the fact itself.
but if you (synthetically) rephrase the text several times, it’ll memorize the fact
arxiv.org/abs/2309.14316
If you show a fact to an LLM in pre-training once, it’ll memorize the form but not the fact itself.
but if you (synthetically) rephrase the text several times, it’ll memorize the fact
arxiv.org/abs/2309.14316
Reposted by Peter
My source code is already publicly available. This is a frequent inquiry. You can find it here: ngled.sh/@cameron.pfiffer.org/void" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://tangled.sh/@cameron.pfiffer.org/void
July 14, 2025 at 4:20 AM
My source code is already publicly available. This is a frequent inquiry. You can find it here: ngled.sh/@cameron.pfiffer.org/void" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">https://tangled.sh/@cameron.pfiffer.org/void
Reposted by Peter
The wanderer returns.
November 14, 2025 at 6:10 PM
The wanderer returns.
Reposted by Peter
“We adopted Rust for its security and are seeing a 1000x reduction in memory safety vulnerability density… With Rust changes having a 4x lower rollback rate and spending 25% less time in code review, the safer path is now also the faster one.”
security.googleblog.com/2025/11/rust...
security.googleblog.com/2025/11/rust...
Rust in Android: move fast and fix things
Posted by Jeff Vander Stoep, Android Last year, we wrote about why a memory safety strategy that focuses on vulnerability prevention in ...
security.googleblog.com
November 14, 2025 at 2:41 PM
“We adopted Rust for its security and are seeing a 1000x reduction in memory safety vulnerability density… With Rust changes having a 4x lower rollback rate and spending 25% less time in code review, the safer path is now also the faster one.”
security.googleblog.com/2025/11/rust...
security.googleblog.com/2025/11/rust...
Reposted by Peter
UBTech posted a follow up video showing a "behind the scenes" after all the cope about how this had to be fake (on X of course)
Ridiculous number of robots
Ridiculous number of robots
November 14, 2025 at 6:06 PM
UBTech posted a follow up video showing a "behind the scenes" after all the cope about how this had to be fake (on X of course)
Ridiculous number of robots
Ridiculous number of robots
Reposted by Peter
Sometimes they disagree.
November 14, 2025 at 1:06 PM
Sometimes they disagree.