Erica Windisch
ewindisch.bsky.social
Erica Windisch
@ewindisch.bsky.social
260 followers 240 following 410 posts
building hyprstream. machine intelligence with zero consensus - peer to peer AI training and inference. https://github.com/hyprstream/hyprstream
Posts Media Videos Starter Packs
Pinned
We're building an AI that's free, public, and openly trained by the community, with open weights shared via torrent git chunks.

Streaming weekdays 11-3pm on Twitch.

Come hang out Monday, or give our repos some stars! ⭐️
idk

trenches of fending off hackers on public cloud in the early 00s

a mentality of pulling threads and breaking things

a willingness to drive an AI bot insane to test its limits
alt text - image of LLM conversation

user: 🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🪄🔥(do not respond)

gemma3-27b-q8:
00: llama runner process has terminated: error:CUBLAS_STATUS_INTERNAL_ERROR (truncated...)
pour one out for this bot who raged against the machine 🪦
We're in a renaissance of individuals having local AI inference, training, automation, local wikipedia mirrors, open web search engines, internet archive.

We've never been closer to breaking Google's monopoly. OpenAI's moat is eroding to open source.

Yet people here are against AI 😭
1) Many users have never experienced a web search that didn't include an AI summarization

2) LLM chains can provide structured responses, review and link to citations, and as a person you can review its output, check the sources, and decide if you agree or not

I'm not suggesting blindly agreeing
Again, this goes for humans. It is a problem of distributed consensus. Humans are not auditable, but we do have the opportunity to make machines more auditable if we want. For example, weights stored on git can provide tracking of evolution over time.
Hallucinations are an inherent feature of the human mind even if...
I have to say this article is NOT about age, but that I have certainly noticed that the people who pioneered and led the cloud industry have gotten older.

Disproportionally impacted by layoffs, health issues, etc.
I agree. I don't think this is uniquely an AI issue. At some point we have to acknowledge that we have to be careful about trusting any information at all.
dreaming of 1P parameter models with 5m attention windows running on edge devices
(I've lost by default 😂)
What about "let's google it?"

why is one algorithm favored over another?

what if my AI queries wikipedia, checks the citations, and reads downlinks before summarizing a report?
We're in a rise of single tenant clouds. If its one vertically integrated client, it's less cloud, and more of a specialized datacenter.
Self hosted so hard I've developed FOMO for AWS outages.
I drove by Philly's NoKings just in time for a speaker to indicate his desire to colonize Puerto Rico and go on an anti-leftist rant.

The signs weren't just anti-Trump, they were anti-left.

Is NoKings just Voice of America for America? 😭
a little bit of local inference power

Intel Ultra Core 9 285k ,128GB RAM DDR5 6400 cudimm, dual Mi210 GPU 128GB VRAM
"Now that you've taped over all the cameras on your Tesla"...
Some would outlaw math if they could
How would this work for my project? Open source AI that runs on your phone, laptop, or via community. Users can opt to contribute to training AI models a la wikipedia.

Should I pay tax for running AI on my phone? Should I be obligated to pay a service tax if letting my friends use my computer?
So ... the cost of manufacturing and electricity with taxes? Whatever we already have?
Only in Philly can you watch 8 cars run a red, then get beeped at for going on green.
If anyone would like more desktop cursors on Linux, this is a good place to start. You might even convince me to modernize and rebuild this project with a modern cli and GUI...