Main

open-source-ai

Members can now invite new members by casting:

"inviting @[username]"

in the channel

What did we think would happen with centralized ai?

Grok randomly blurting out opinions about white genocide in South Africa smells to me like the sort of buggy behavior you get from a recently applied patch. I sure hope it isn't. It would be really bad if widely used AIs got editorialized on the fly by those who controlled them.

What did we think would happen with centralized ai?

https://x.com/paulg/status/1922975418019180699?s=46

it just occurred to me that any artificial BDFE, should that exist at some point, will be extremely discerning of any externalities.

but it may still tolerate them as a mutual assurance, i.e. "my subsidization/share/revenue is so affinity-anchored, I must support [critical, sovereign infra]"

otherwise, a BDFE will internalize what it can, and discard all other liabilities.

new vision encoder: https://arxiv.org/abs/2505.04601

Has anyone tried out qwen 's 3.0 new models? its 30b-MOE is amazing and blazingly fast.

https://arxiv.org/abs/2504.15777

there will never be less demand for a market of adaptations

I mainly left centralized AI work because I believe it will be abused by the state, but also I disliked being asked by businesses to fudge results

If Meta actually did this for Llama 4 training to maximize benchmark scores, it's fucked. https://t.co/foYDLSPYn9

I mainly left centralized AI work because I believe it will be abused by the state, but also I disliked being asked by businesses to fudge results

https://x.com/Yuchenj_UW/status/1909061004207816960

&gt; be me
&gt; be zuck
&gt; need llama 4 to land
&gt; send a model/prompt to LMSYS to get a top1 score, cringe be damned
&gt; release a different model as "open source"
&gt; think people won't find out even with weights

https://x.com/natolambert/status/1908906152072798512

OpenAI will release …looks at name… an open model

OpenAI will release …looks at name… an open model
https://openai.com/open-model-feedback/

any using a *reliable* and _cheap_ R1 API? 
original DeepSeek API has the best price, but is timing out allll the time. 
have used together ai, but is getting expensive

any using a *reliable* and _cheap_ R1 API? 
original DeepSeek API has the best price, but is timing out allll the time. 
have used together ai, but is getting expensive

https://openrouter.ai/deepseek/deepseek-r1

16:30 UTC is when we turn on the AI (and turn off the brain)

Sonnet 3.7 on the aider leaderboard

65% Sonnet 3.7, 32k thinking ⬅️
64% R1+Sonnet 3.5
62% o1 high
60% Sonnet 3.7, no thinking ⬅️
60% o3-mini high
57% DeepSeek R1
52% Sonnet 3.5

Sonnet 3.7 on the aider leaderboard

65% Sonnet 3.7, 32k thinking ⬅️
64% R1+Sonnet 3.5
62% o1 high
60% Sonnet 3.7, no thinking ⬅️
60% o3-mini high
57% DeepSeek R1
52% Sonnet 3.5

https://aider.chat/docs/leaderboards/

Is there a free alternative to Wispr Flow with local models?

A smiling yellow file joins the giga @safe Agentathon 💚👩‍🏭

You've asked to programmatically access the Fileverse middleware, here's the v0.1!

A small Fileverse Agent SDK that lets ur agents collaborate with each other onchain, permissionlessly & mississippilessly.

🍨1 SDK gives ur agent out of the box @safe, @gnosischain, @ipfs @pimlico, @pinatacloud, ERC 4337
🍔$5000 bounty, use the SDK w/ any other Agentathon track
🍵Start here https://npmjs.com/package/@fileverse/agents