🚨LAUNCH NEWS🚨 We are about to experience a golden age of open source AI as a response to many companies losing access to AI APIs and the combined inference bills hitting tens of billions of dollars. There are now more than 100,000 models uploaded to huggingface every month, but small task specific models are still hard to discover and running a mixed model workload across pipelines and real-time queries is a challenge. Building on our heritage of search & document processing across e-com and enterprise, we decided to *BET OUR WHOLE COMPANY* on a new product that launches today: a self-hostable open source cluster we call the Superlinked Inference Engine. Help us make open source AI viable by starring, contributing and using SIE on your own hardware in the safety of your own cloud *for free* 🙏 Repo link and more info: superlinked.com/blog/launch Onwards! PS: We would not be able to do this without our wonderful partners @trychroma @lancedb @qdrant_engine @weaviate_io - more on this coming soon!

Heard of a top law firm completely cutting off Harvey and all API calls to Claude/OpenAI after they saw what openAI did with navier strokes Now hosting everything locally with GPUs they own, running Kimi and deepseek Think you’ll see a lot more of this going forward

Today's world runs on security through obscurity, minus the obscurity.

you are levered on platform risk if you are an OAI / Anth-only shop

Hanging out with p/d disaggregator & KV cache compression enjoyers in Santa Clara tomorrow evening, the table has ~2-3 more seats available - anybody keen? Yes we will discuss how to abliterate a model so you can use it for cybersec work ^_^

Yo @sedatesnail there definitely are emotions / trauma that are not hormetic.. you can observe the toxicity build up as a chronic stress response that gets worse / solidifies with exposure. I liked your essay, cool framework :-)

With the rise of #Jev, OpenAI's #DecisionAPI and a flurry of Open Source Decision Models coming out recently, knowing when your agent should think fast and when it should think slow is going to be one of the biggest design decisions in agentic AI. With the rise of System 1 models, you no longer have to send every step through System 2 thinking. The question is which steps belong where. But we've built decision models before! So what's new this time? Daniel Svonava (@svonava) and I hosted a small group of AI engineers working across finance, legal, healthcare and more to dig into this. Here's when System 1 makes sense: - You need decisions, not long text. - You need them fast. - You have labeled data to evaluate models and define policy. The big unlock is that open source models can now give you fast outputs with LLM-grade decision-making. All you need is a few use case examples to fine-tune an open source decision model, and infrastructure built to compute the forward pass fast. Here's when open source System 1 makes sense: - You want to trust your data to a black-box model. - You have enough data to fine-tune an open weights model. - You have the infra, and your model needs to sit next to your agent for speed. Lots to build here. Try them with @superlinked superlinked.com/

As a founder, if you don’t have an audience of your own, you will be misrepresented in media and there won’t be anything you can do about it. Learning: get on that social media grind!

Me and @RahulParundekar in mission bay tomorrow giving the 201 on decision models with free pizza and drinks - come hang out! luma.com/q6ytgxe2

Very cool SIE learning resource from @theNeomatrix369 I just saw fly around in our discord: neomatrix369.github.io/demos… Mani uses SIE to power AgentVetter, his security scanning agent - worth checking out, I think security is the next killer app for AI :-)

It’s Open Source AI Week all over the Bay Area next week! With more and more open-source System 1 models coming out, I wanted to get the people working with them into one room. Why System 1? Most agents think slow: they send every yes/no call to a big, slow LLM. Sometimes that’s warranted, like when the model needs to reason through an answer or explain it. System 1 models are the other half. These are small open-source models (e.g. GLiClass, GLiNER, Laya) that answer in milliseconds. Running one on your laptop is one thing, but serving them at scale is a whole different challenge. @svonava and I are hosting an evening on when fast is good enough, when you need slow, and how to hand off between the two. There’ll be talks, a discussion, and then beer at Fieldworks. Here’s the Luma: luma.com/q6ytgxe2 #OpenSourceAIWeek #OpenSourceAI #AIEngineering

Today was hard to step in the room for the ai conference after reading that @AnitaB_org has been shut down The gender imbalance was even more pressing now that the one conference for women in STEM shut down because of lack of funding I’m no Melinda Gates to save this with money on tap, but I promise I will do something when I’m back in SF I will do my share of building community and belonging for women in tech. The size of your wallet does not have to dictate the size of your impact And I swear my events won’t be all male

messages like these from our customers just make my day @usedayly 🥰

Love building with ML - and getting a room full of developers excited about it? We're hiring a DevRel Engineer at Superlinked to grow the community around our open source Inference Engine. You'll speak at conferences, run workshops and hackathons, build demos worth trying, and turn developer conversations into a better product. We're looking for someone who can ship code, tell a clear technical story, and bring people together, with a strong focus on the SF AI/ML scene. Sound like you? Come join us! jobs.superlinked.com/jobs/85… #hiring #devrel

So, squares huh? @ojoshe @UniqueDima @naveen23k

today in @softwaredoug class on search talking about small model inference, showing how a reranker can be used as a model router with github.com/superlinked/sie

Some goodies for the encoder models, past few releases have been focused on these, you can check more at superlinked.com/docs/referen…

Ya que termino el verano, comento por acá lo que estuve estudiando. El caso es que todos hablan de agentes, Rags y tal, pero al menos en español, poco se habla de lo que esta atras, la inferencia! Pa' eso me estuve creando este proyecto, la infra de un coding agent en k8s