ai benchmarks are funny because metr and ai futures project have gone to make specific benchmarks something like objects of religious devotion when they're simply a signal that goodharts out within 1-3 months of introduction every time. this has value but it's transient value
November 1, 2026 at 8:07 PM
ai benchmarks are funny because metr and ai futures project have gone to make specific benchmarks something like objects of religious devotion when they're simply a signal that goodharts out within 1-3 months of introduction every time. this has value but it's transient value
san francisco is so stupid why does artificial analysis have ads on Muni showing benchmarks
November 1, 2026 at 3:51 AM
san francisco is so stupid why does artificial analysis have ads on Muni showing benchmarks
AI should benefit everyone. So we built a benchmark.
That’s not how benchmarks work.
That’s just Goodhart’s law as a service.
That’s not how benchmarks work.
That’s just Goodhart’s law as a service.
For AGI to benefit all of humanity, it should help people with the everyday tasks that matter most to them.
November 2, 2026 at 6:18 PM
AI should benefit everyone. So we built a benchmark.
That’s not how benchmarks work.
That’s just Goodhart’s law as a service.
That’s not how benchmarks work.
That’s just Goodhart’s law as a service.
We’re releasing new DeepSeek-V4.1-Flash-1017 GGUFs, ranking highest across benchmarks.
Dynamic V3.1 leads across nearly all benchmarks: GPQA Diamond, AIME25, LiveCodeBench v6, MMLU-Pro, Aider & more.
Run on 128GB or 64GB+ RAM.
GGUF: huggingface.co/unsloth/Deep...
Blog: unsloth.ai/docs/models/...
Dynamic V3.1 leads across nearly all benchmarks: GPQA Diamond, AIME25, LiveCodeBench v6, MMLU-Pro, Aider & more.
Run on 128GB or 64GB+ RAM.
GGUF: huggingface.co/unsloth/Deep...
Blog: unsloth.ai/docs/models/...
October 30, 2026 at 1:14 PM
We’re releasing new DeepSeek-V4.1-Flash-1017 GGUFs, ranking highest across benchmarks.
Dynamic V3.1 leads across nearly all benchmarks: GPQA Diamond, AIME25, LiveCodeBench v6, MMLU-Pro, Aider & more.
Run on 128GB or 64GB+ RAM.
GGUF: huggingface.co/unsloth/Deep...
Blog: unsloth.ai/docs/models/...
Dynamic V3.1 leads across nearly all benchmarks: GPQA Diamond, AIME25, LiveCodeBench v6, MMLU-Pro, Aider & more.
Run on 128GB or 64GB+ RAM.
GGUF: huggingface.co/unsloth/Deep...
Blog: unsloth.ai/docs/models/...
How good is your kobold's endurance? Leave your benchmarks in the replies!
Day 30 - Time
#Boldtober2026
Day 30 - Time
#Boldtober2026
October 30, 2026 at 9:44 PM
How good is your kobold's endurance? Leave your benchmarks in the replies!
Day 30 - Time
#Boldtober2026
Day 30 - Time
#Boldtober2026
watching the benchmarks again after the expacs is so funny. Ok so none of that happened (being obtuse on purpose)
November 2, 2026 at 9:37 PM
watching the benchmarks again after the expacs is so funny. Ok so none of that happened (being obtuse on purpose)
i imagine itll be obvious to anyone who knows ball how it's going based on benchmarks elsewhere, guess we'll just have to loudly celebrate those benchmarks being met or exceeded
October 31, 2026 at 7:04 PM
i imagine itll be obvious to anyone who knows ball how it's going based on benchmarks elsewhere, guess we'll just have to loudly celebrate those benchmarks being met or exceeded
📊 Can you really trust your benchmarks?
A big thank you to Kemal Akkoyun for showing us how to avoid misleading results, measure Go performance accurately, and catch regressions before they reach production.
Better benchmarks, better Go! 🚀 #GoLab2026
A big thank you to Kemal Akkoyun for showing us how to avoid misleading results, measure Go performance accurately, and catch regressions before they reach production.
Better benchmarks, better Go! 🚀 #GoLab2026
November 3, 2026 at 10:39 AM
📊 Can you really trust your benchmarks?
A big thank you to Kemal Akkoyun for showing us how to avoid misleading results, measure Go performance accurately, and catch regressions before they reach production.
Better benchmarks, better Go! 🚀 #GoLab2026
A big thank you to Kemal Akkoyun for showing us how to avoid misleading results, measure Go performance accurately, and catch regressions before they reach production.
Better benchmarks, better Go! 🚀 #GoLab2026
These hucksters are hawking their fake benchmarks on Hacker News. Fake because performance numbers DON'T COUNT IF YOUR CODE IS BROKEN JESUS CHRIST
![nikez96 5 hours ago [dead] | parent | context | next [–] | on: Jepsen: ParticleDB 1.1.4
https://particledb.ai/benchmarks/key-value/ You can find the details on redis vs particledb benchmark that anyone can run for themselves
nikez96 7 hours ago | parent | context | prev [–] | on: Jepsen: ParticleDB 1.1.4
I'm part of the ParticleDB team. We took these findings](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:sl3g3nhnad34sfhjnqdivbzd/bafkreihkoyfkmjmmwlcg24kbgubluth6p7ch2tvlqiu2uy4ryzobebhyja)
November 1, 2026 at 3:20 AM
These hucksters are hawking their fake benchmarks on Hacker News. Fake because performance numbers DON'T COUNT IF YOUR CODE IS BROKEN JESUS CHRIST
I think the last time benchmarks had meaningful value for the average person might have been eight months ago. I get mad when people gork it to AA or DeepSWE now. HumanRelevantBench is saturated
ai benchmarks are funny because metr and ai futures project have gone to make specific benchmarks something like objects of religious devotion when they're simply a signal that goodharts out within 1-3 months of introduction every time. this has value but it's transient value
November 1, 2026 at 10:35 PM
I think the last time benchmarks had meaningful value for the average person might have been eight months ago. I get mad when people gork it to AA or DeepSWE now. HumanRelevantBench is saturated
VERY helpful graphic by Drew for following VA01 and VA05 results tonight! bluevirginia.us/2026/11/virg...
November 3, 2026 at 4:41 PM
VERY helpful graphic by Drew for following VA01 and VA05 results tonight! bluevirginia.us/2026/11/virg...
Some part of me still can’t believe that so many people and institutions have decided to not do their jobs with a baseline degree of care or like meeting basic benchmarks? Like we looked around at how nothing works and doubled down.
I understand there are hiccups in every new technology. Until AI works perfectly it should NOT be dealing with medical patient care/customer service.
This isn’t an “oopsie your HBO Max has ads when you’re paying for no ads.”
It’s serious health stuff! Fuck off!!!
This isn’t an “oopsie your HBO Max has ads when you’re paying for no ads.”
It’s serious health stuff! Fuck off!!!
I’ve now experienced AI fucking up TWO medical things for me in the past month.
1. Sent me the wrong summary of a bad blood test result via patient portal that scared the shit out of me
2. An AI agent (who I did not know was an AI agent) called me today to schedule a test my doctor did not order
1. Sent me the wrong summary of a bad blood test result via patient portal that scared the shit out of me
2. An AI agent (who I did not know was an AI agent) called me today to schedule a test my doctor did not order
November 2, 2026 at 11:26 PM
Some part of me still can’t believe that so many people and institutions have decided to not do their jobs with a baseline degree of care or like meeting basic benchmarks? Like we looked around at how nothing works and doubled down.
HAVN HS 360 case review: www.youtube.com/watch?v=OgHT...
Fixing the Vertical GPU Problem | HAVN HS 360 Case Review, Thermals, & Benchmarks
YouTube video by Gamers Nexus
www.youtube.com
October 29, 2026 at 3:04 PM
HAVN HS 360 case review: www.youtube.com/watch?v=OgHT...
We’re hitting our benchmarks in Travis County and Georgia (probably Dallas, too)
It’s now past 9PM, one hour (plus time to clear lines). Here’s how things stand in Travis *so far*. I can’t compare to 9PM in 2024 unfortunately.
2024, total: 47,640
2026, total SO FAR: 46,703 (98.0%)
Stay tuned for the post-10PM final update!
2024, total: 47,640
2026, total SO FAR: 46,703 (98.0%)
Stay tuned for the post-10PM final update!
October 31, 2026 at 2:14 AM
We’re hitting our benchmarks in Travis County and Georgia (probably Dallas, too)
according to Morris, gop is NOT hitting their benchmarks in Florida that they need to be hitting, and the early hours tend to be when the gop does the best. Day's still young, though.
November 3, 2026 at 2:46 PM
according to Morris, gop is NOT hitting their benchmarks in Florida that they need to be hitting, and the early hours tend to be when the gop does the best. Day's still young, though.
One early house race to watch will be FL-7. If Democrats have a really good night, they could flip this Trump+12 district in Central Florida. Here are the benchmarks Democrats have to hit in Seminole & Volusia County.
LINK: electioncord.com/permutator/
LINK: electioncord.com/permutator/
November 2, 2026 at 11:29 PM
One early house race to watch will be FL-7. If Democrats have a really good night, they could flip this Trump+12 district in Central Florida. Here are the benchmarks Democrats have to hit in Seminole & Volusia County.
LINK: electioncord.com/permutator/
LINK: electioncord.com/permutator/
Over on LinkedIn, I saw folks discussing how important it was to have new benchmarks in immigration law + AI, re: a prominent immigration tech nonprofit announcing a partnership with Anthropic.
And like...sure? But let's start with basic benchmarks on case outcomes, representation rates, fees...
And like...sure? But let's start with basic benchmarks on case outcomes, representation rates, fees...
October 30, 2026 at 10:01 PM
Over on LinkedIn, I saw folks discussing how important it was to have new benchmarks in immigration law + AI, re: a prominent immigration tech nonprofit announcing a partnership with Anthropic.
And like...sure? But let's start with basic benchmarks on case outcomes, representation rates, fees...
And like...sure? But let's start with basic benchmarks on case outcomes, representation rates, fees...
Booted up Tokon to make sure my shit was dialed in.
Did you know, benchmarks aren't just a good idea to make sure everyone is playing in a stable environment, it's also good so users can fine tune their settings while getting the results they want.
I got it looking pretty nice, too.
Did you know, benchmarks aren't just a good idea to make sure everyone is playing in a stable environment, it's also good so users can fine tune their settings while getting the results they want.
I got it looking pretty nice, too.
November 3, 2026 at 5:42 AM
Booted up Tokon to make sure my shit was dialed in.
Did you know, benchmarks aren't just a good idea to make sure everyone is playing in a stable environment, it's also good so users can fine tune their settings while getting the results they want.
I got it looking pretty nice, too.
Did you know, benchmarks aren't just a good idea to make sure everyone is playing in a stable environment, it's also good so users can fine tune their settings while getting the results they want.
I got it looking pretty nice, too.
Private benchmarks often show open models are further behind, but the problem with private benchmarks is you don’t have the community to squash the inevitable bugs for you. I'm 50/50 on if I think private or open-source is the best approach.
October 28, 2026 at 1:35 PM
Private benchmarks often show open models are further behind, but the problem with private benchmarks is you don’t have the community to squash the inevitable bugs for you. I'm 50/50 on if I think private or open-source is the best approach.
China just released a tiny model this week that crushes previous long-form OCR benchmarks. My goats. I will be unemployed in 4 years.
November 1, 2026 at 7:12 PM
China just released a tiny model this week that crushes previous long-form OCR benchmarks. My goats. I will be unemployed in 4 years.
listen, I just decide which models are good based purely on vibes, and if a benchmark agrees then that's a good benchmarks. no need to make it any more complicated.
November 1, 2026 at 10:56 PM
listen, I just decide which models are good based purely on vibes, and if a benchmark agrees then that's a good benchmarks. no need to make it any more complicated.
Benchmarks (and other basic info) about Alabama's 2nd Congressional District, which Crystal Ball has as a tossup!
LINK: docs.google.com/spreadsheets...
LINK: docs.google.com/spreadsheets...
November 1, 2026 at 1:48 AM
Benchmarks (and other basic info) about Alabama's 2nd Congressional District, which Crystal Ball has as a tossup!
LINK: docs.google.com/spreadsheets...
LINK: docs.google.com/spreadsheets...
Memdoor: How Five AI Models Working Together Beat Frontier Models on Code Benchmarks
The Power of Collaboration Over Scale
In the rapidly evolving landscape of AI-assisted programming, a surprising discovery chall…
#ai #machine-learning #programming #open-source #workflow #coding #benchmarks
The Power of Collaboration Over Scale
In the rapidly evolving landscape of AI-assisted programming, a surprising discovery chall…
#ai #machine-learning #programming #open-source #workflow #coding #benchmarks
Memdoor: How Five AI Models Working Together Beat Frontier Models on Code Benchmarks
The Power of Collaboration Over Scale
In the rapidly evolving landscape of AI-assisted programming, a surprising discovery challenges the conventional wisdom that bigger models are always better. Memdoor, an open-source project by guregodevo, demonst...
techplanet.today
November 3, 2026 at 10:22 AM
Memdoor: How Five AI Models Working Together Beat Frontier Models on Code Benchmarks
The Power of Collaboration Over Scale
In the rapidly evolving landscape of AI-assisted programming, a surprising discovery chall…
#ai #machine-learning #programming #open-source #workflow #coding #benchmarks
The Power of Collaboration Over Scale
In the rapidly evolving landscape of AI-assisted programming, a surprising discovery chall…
#ai #machine-learning #programming #open-source #workflow #coding #benchmarks