Peter's Avatar

Peter

@petmouse.bluesky.mousses.xyz

I like computers and nature ๐Ÿจโ˜ฆ๏ธ

75 Followers  |  68 Following  |  169 Posts  |  Joined: 25.01.2025  |  2.2476

Latest posts by petmouse.bluesky.mousses.xyz on Bluesky

I just saw companion. Great movie and a very realistic cautionary tale, but I can't help but think were already there with 4o

23.11.2025 03:34 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

Nondeterminism is always fun

22.11.2025 23:28 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0

I have, in fact seen private email. It involves sending an encrypted attachment containing the actual message

22.11.2025 22:48 โ€” ๐Ÿ‘ 1    ๐Ÿ” 0    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0

This is why I have stopped using Gmail except as the email that I give when online shopping. My main email is Proton now.

Google turned their service into a dumping ground for spam.

So that's how I use it.

22.11.2025 01:12 โ€” ๐Ÿ‘ 164    ๐Ÿ” 30    ๐Ÿ’ฌ 13    ๐Ÿ“Œ 1
A small tabby and white kitten and black bird, likely a jackdaw, standing together in a dirt yard, looking to the left. They seem relaxed and at ease with each other.

A small tabby and white kitten and black bird, likely a jackdaw, standing together in a dirt yard, looking to the left. They seem relaxed and at ease with each other.

The same bird and kitten from the previous photo, this time strolling away from the camera in a different outdoor setting together, side by side.

The same bird and kitten from the previous photo, this time strolling away from the camera in a different outdoor setting together, side by side.

This is one of my favorite sets in my collection. Somewhere out there in the 1960s, a kitten and bird were pals. ๐Ÿ’•

21.11.2025 18:40 โ€” ๐Ÿ‘ 1180    ๐Ÿ” 159    ๐Ÿ’ฌ 16    ๐Ÿ“Œ 5
an alien looking orange kitten with a googly eye placed in between his eyes on his forehead

an alien looking orange kitten with a googly eye placed in between his eyes on his forehead

he sees all

21.11.2025 13:15 โ€” ๐Ÿ‘ 1872    ๐Ÿ” 354    ๐Ÿ’ฌ 26    ๐Ÿ“Œ 35

Their real fear is probably that they lose the ability to use their data extraction tools with most Android and iOS devices rather than a tiny minority of GrapheneOS devices. Despite how frustrated we are with Google, we're seriously considering trying to help them a bit more because of this.

21.11.2025 06:54 โ€” ๐Ÿ‘ 14    ๐Ÿ” 3    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0

Evidently Nano Banana Pro has some surprising gaps in its guardrails.

20.11.2025 22:14 โ€” ๐Ÿ‘ 3    ๐Ÿ” 1    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0
The "know the work rules" meme with 

"we made an AI obsessed with the golden gate bridge"

vs

"we made an AI obsessed with Elon Musk"

The "know the work rules" meme with "we made an AI obsessed with the golden gate bridge" vs "we made an AI obsessed with Elon Musk"

20.11.2025 21:42 โ€” ๐Ÿ‘ 61    ๐Ÿ” 12    ๐Ÿ’ฌ 2    ๐Ÿ“Œ 0
Calico kitten laying upside down, looking like an adorable little goblin

Calico kitten laying upside down, looking like an adorable little goblin

Calico kitten licking her paw, close up fish eye shot making her nose and tongue look wide

Calico kitten licking her paw, close up fish eye shot making her nose and tongue look wide

21.11.2025 05:30 โ€” ๐Ÿ‘ 0    ๐Ÿ” 0    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Video thumbnail

Figure 02 working in BMW plant. they did this every day for months and contributed to the production of tens of thousands of cars.

20.11.2025 18:32 โ€” ๐Ÿ‘ 8    ๐Ÿ” 2    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0

I really hope signal for Android will get material 3 expressive just so that their Android app isn't getting neglected with the bare minimum. If they are aiming for a native look for IOS then it would only make sense to do the same by implementing a native look for Android.

21.11.2025 01:07 โ€” ๐Ÿ‘ 0    ๐Ÿ” 1    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Post image Post image Post image Post image

This is a Nano Banana Pro appreciation thread.

20.11.2025 18:45 โ€” ๐Ÿ‘ 8    ๐Ÿ” 2    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 1
Preview
Nano Banana Pro aka gemini-3-pro-image-preview is the best available image generation model Hot on the heels of Tuesdayโ€™s Gemini 3 Pro release, today itโ€™s Nano Banana Pro, also known as Gemini 3 Pro Image. Iโ€™ve had a few days of preview access โ€ฆ

Nano Banana Pro, released this morning, is clearly the best image generation model. Superb instruction following, plus it can generate full infographics (with correct spelling and properly rendered text!) from a short prompt based on running extra searches simonwillison.net/2025/Nov/20/...

20.11.2025 16:34 โ€” ๐Ÿ‘ 162    ๐Ÿ” 23    ๐Ÿ’ฌ 11    ๐Ÿ“Œ 7
Video thumbnail

โ€œA harvard professor named in the epstein files has now resigned. Accountability is finally catching up with institutions that protected predators for decades.โ€ Hart โ€˜28 Dem Pursuing.com en.wikipedia.org/wiki/Lawrenc... ๐Ÿ‡บ๐Ÿ‡ธ๐Ÿ™

20.11.2025 14:21 โ€” ๐Ÿ‘ 249    ๐Ÿ” 69    ๐Ÿ’ฌ 19    ๐Ÿ“Œ 14

The cat has once again deployed a devastating DDoS attack against my "lap" microservice, causing me to miss work,

20.11.2025 11:36 โ€” ๐Ÿ‘ 53    ๐Ÿ” 7    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Post image

โ€˜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.โ€˜ garymarcus.substack.com/p/hot-take-o... by @garymarcus.bsky.social

19.11.2025 17:16 โ€” ๐Ÿ‘ 72    ๐Ÿ” 20    ๐Ÿ’ฌ 2    ๐Ÿ“Œ 2

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.

19.11.2025 14:50 โ€” ๐Ÿ‘ 93    ๐Ÿ” 16    ๐Ÿ’ฌ 5    ๐Ÿ“Œ 0
Ignas | DeFi @Defilgnas
X.com
Bitcoin is being dumped by long-term holders due to quantum risk. A quantum computer with sufficient logical Qubits will be able to break the private keys of 20-30% of the coins. This means that Bitcoin is effectively providing a US$500 billion 'science prize' to the first company to build the computer which breaks the encryption.
There are some potential solutions for the problem, but it is clear that because Bitcoin has no management or leaders, they won't be adopted. The Bitcoin protocol will most likely die once quantum hacks starts happening, as confidence will be destroyed.

Ignas | DeFi @Defilgnas X.com Bitcoin is being dumped by long-term holders due to quantum risk. A quantum computer with sufficient logical Qubits will be able to break the private keys of 20-30% of the coins. This means that Bitcoin is effectively providing a US$500 billion 'science prize' to the first company to build the computer which breaks the encryption. There are some potential solutions for the problem, but it is clear that because Bitcoin has no management or leaders, they won't be adopted. The Bitcoin protocol will most likely die once quantum hacks starts happening, as confidence will be destroyed.

seems bad

19.11.2025 15:08 โ€” ๐Ÿ‘ 45    ๐Ÿ” 6    ๐Ÿ’ฌ 17    ๐Ÿ“Œ 12
Preview
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.

On November 18 Cloudflare experienced a service outage. Here's a detailed breakdown of what happened. https://cfl.re/43Bw8AI

18.11.2025 23:49 โ€” ๐Ÿ‘ 104    ๐Ÿ” 61    ๐Ÿ’ฌ 6    ๐Ÿ“Œ 26

software engineering has taught me that it's a lot easier to withstand a nuclear attack than a fat-fingered config

19.11.2025 00:06 โ€” ๐Ÿ‘ 8    ๐Ÿ” 1    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 1
A dark-themed scatterplot titled โ€œARC-AGI-2 LEADERBOARD.โ€

Axes
	โ€ข	Y-axis: Score (%), ranging from 0% up to 55%.
	โ€ข	X-axis: Cost per task ($), logarithmic from $1eโ€“3 to $1K.

Legend

Top-left: ARC Prize | Verified

Data Points

Colorful triangular markers represent many different model variants. The plot shows a strong costโ€“performance frontier: most points cluster below 10% score, but a few models stand far above.

โธป

High-scoring outliers (upper-right and upper-mid):
	โ€ข	Gemini 3 Deep Think (Preview) โ€” ~50% score, very high cost (โ‰ˆ$500โ€“$1000).
Green triangle on the far right and highest on the chart.
	โ€ข	Gemini 3 Pro โ€” ~30% score, cost around $1.
Green triangle centered horizontally.
	โ€ข	GPT-5 Pro โ€” ~17% score, cost โ‰ˆ**$10**.
Blue triangle above the mid-cost range.
	โ€ข	Grok 4 (Thinking) โ€” ~14โ€“15% score, cost around $1โ€“$2.
Pink triangle slightly below GPT-5 Pro.
	โ€ข	Claude Sonnet 4.5 (Thinking 32K) โ€” ~13% score, cost โ‰ˆ**$1**.
Red triangle near Grok 4.
	โ€ข	GPT-5 (High) โ€” ~10โ€“12% score, cost around $1.
Blue triangle.
	โ€ข	o3 (High) โ€” around 8โ€“9%, cost โ‰ˆ**$0.50**.
Blue triangle just below GPT-5 (High).

โธป

Mid-range cluster: 1โ€“7% score

Dozens of small colored triangles (various models and reasoning modes) cluster in the low-cost region (0.01โ€“1 dollars). This includes:
	โ€ข	o3-mini (High)
	โ€ข	o3 (Medium)
	โ€ข	GPT-4.5
	โ€ข	o3-mini (Medium)
	โ€ข	GPT-5 Mini (various modes)
	โ€ข	Tiny/low-cost models from multiple families

These fill the bottom-left quadrant.

โธป

Very low-cost, very low-score models

At the far-left (cost ~$1eโ€“3 to 1eโ€“2), many points sit at 0โ€“2%, representing extremely cheap but weak models.

โธป

Overall picture
	โ€ข	Gemini models dominate the top (Gemini 3 Deep Think โ†’ ~50%, Gemini 3 Pro โ†’ ~30%).
	โ€ข	No other model exceeds ~17%.
	โ€ข	Cost rises steeply with performanceโ€”especially for the top Gemini variants.

A dark-themed scatterplot titled โ€œARC-AGI-2 LEADERBOARD.โ€ Axes โ€ข Y-axis: Score (%), ranging from 0% up to 55%. โ€ข X-axis: Cost per task ($), logarithmic from $1eโ€“3 to $1K. Legend Top-left: ARC Prize | Verified Data Points Colorful triangular markers represent many different model variants. The plot shows a strong costโ€“performance frontier: most points cluster below 10% score, but a few models stand far above. โธป High-scoring outliers (upper-right and upper-mid): โ€ข Gemini 3 Deep Think (Preview) โ€” ~50% score, very high cost (โ‰ˆ$500โ€“$1000). Green triangle on the far right and highest on the chart. โ€ข Gemini 3 Pro โ€” ~30% score, cost around $1. Green triangle centered horizontally. โ€ข GPT-5 Pro โ€” ~17% score, cost โ‰ˆ**$10**. Blue triangle above the mid-cost range. โ€ข Grok 4 (Thinking) โ€” ~14โ€“15% score, cost around $1โ€“$2. Pink triangle slightly below GPT-5 Pro. โ€ข Claude Sonnet 4.5 (Thinking 32K) โ€” ~13% score, cost โ‰ˆ**$1**. Red triangle near Grok 4. โ€ข GPT-5 (High) โ€” ~10โ€“12% score, cost around $1. Blue triangle. โ€ข o3 (High) โ€” around 8โ€“9%, cost โ‰ˆ**$0.50**. Blue triangle just below GPT-5 (High). โธป Mid-range cluster: 1โ€“7% score Dozens of small colored triangles (various models and reasoning modes) cluster in the low-cost region (0.01โ€“1 dollars). This includes: โ€ข o3-mini (High) โ€ข o3 (Medium) โ€ข GPT-4.5 โ€ข o3-mini (Medium) โ€ข GPT-5 Mini (various modes) โ€ข Tiny/low-cost models from multiple families These fill the bottom-left quadrant. โธป Very low-cost, very low-score models At the far-left (cost ~$1eโ€“3 to 1eโ€“2), many points sit at 0โ€“2%, representing extremely cheap but weak models. โธป Overall picture โ€ข Gemini models dominate the top (Gemini 3 Deep Think โ†’ ~50%, Gemini 3 Pro โ†’ ~30%). โ€ข No other model exceeds ~17%. โ€ข Cost rises steeply with performanceโ€”especially for the top Gemini variants.

Gemini 3 DeepThink is 2x SOTA on ARC-AGI-2

18.11.2025 20:59 โ€” ๐Ÿ‘ 27    ๐Ÿ” 3    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0
Oriol Vinyals &
โ€ข @OriolVinyalsML โ€ข2h
The secret behind Gemini 3?
Simple: Improving pre-training & post-training
Pre-training: Contra the popular belief that scaling is overโ€”which we discussed in our NeurIPS '25 talk with @ilyasut and @quocleixโ€” the team delivered a drastic jump. The delta between 2.5 and 3.0 is as big as we've ever seen. No walls in sight!
Post-training: Still a total greenfield. There's lots of room for algorithmic progress and improvement, and 3.0 hasn't been an exception, thanks to our stellar team.
Congratulations to the whole team

Oriol Vinyals & โ€ข @OriolVinyalsML โ€ข2h The secret behind Gemini 3? Simple: Improving pre-training & post-training Pre-training: Contra the popular belief that scaling is overโ€”which we discussed in our NeurIPS '25 talk with @ilyasut and @quocleixโ€” the team delivered a drastic jump. The delta between 2.5 and 3.0 is as big as we've ever seen. No walls in sight! Post-training: Still a total greenfield. There's lots of room for algorithmic progress and improvement, and 3.0 hasn't been an exception, thanks to our stellar team. Congratulations to the whole team

Gemini 3 is indeed a much larger model

both pre-training (model size) & post-training (RL) scaling

18.11.2025 21:21 โ€” ๐Ÿ‘ 23    ๐Ÿ” 2    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 2
The myth of consensual internet.
Browser: I consent!
Host: I consent!
Cloudflare: I don't!
Isn't there somebody you forgot to ask?

The myth of consensual internet. Browser: I consent! Host: I consent! Cloudflare: I don't! Isn't there somebody you forgot to ask?

19.11.2025 00:05 โ€” ๐Ÿ‘ 126    ๐Ÿ” 30    ๐Ÿ’ฌ 2    ๐Ÿ“Œ 0
Preview
Two Weeks of Surveillance Footage From ICE Detention Center โ€˜Irretrievably Destroyedโ€™ "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."

New: DHS claimed in court proceedings that nearly two weeks worth of surveillance footage from ICEโ€™s Broadview Detention Center in suburban Chicago has been โ€œirretrievably destroyedโ€ and may not be able to be recovered. ACLU of Illinois said "Hoping DHS can explain"

www.404media.co/two-weeks-of...

18.11.2025 17:03 โ€” ๐Ÿ‘ 181    ๐Ÿ” 80    ๐Ÿ’ฌ 5    ๐Ÿ“Œ 15

Jeff selling shirts might be the new Pizza Index but for internet problems

18.11.2025 13:52 โ€” ๐Ÿ‘ 30    ๐Ÿ” 2    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 0
Preview
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.

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.

18.11.2025 15:44 โ€” ๐Ÿ‘ 111    ๐Ÿ” 54    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 7
Post image

Announcing RefCOCO-M, a refreshed RefCOCO with pixel-accurate masks and the problematic prompts removed. Better data for better evaluation.

huggingface.co/datasets/moo...

18.11.2025 06:55 โ€” ๐Ÿ‘ 10    ๐Ÿ” 2    ๐Ÿ’ฌ 1    ๐Ÿ“Œ 0

lol he sounds like the fucking Penguin here

18.11.2025 03:16 โ€” ๐Ÿ‘ 2214    ๐Ÿ” 213    ๐Ÿ’ฌ 108    ๐Ÿ“Œ 8
Preview
Announcing the Agentic Learning SDK: Add state to anything This is huge for adoption - it removes the biggest friction point for teams wanting Lettaโ€™s memory capabilities without refactoring existing applications. Key Insight: The context manager pattern (w...

Oh thanks for the reminder. We released the learning SDK to add state to ~whatever, looking for feedback:

forum.letta.com/t/announcing...

16.11.2025 19:24 โ€” ๐Ÿ‘ 8    ๐Ÿ” 2    ๐Ÿ’ฌ 0    ๐Ÿ“Œ 1

@petmouse.bluesky.mousses.xyz is following 20 prominent accounts