Lukas Wesemann

Building local AI, Co-founder @basecmpt

[Lukas Wesemann @lukasletsgo]

@TheAhmadOsman fastest runtime on apple silicon, 78% faster on prefill for 30B class MOEs : @basecmpt

[Base Compute @basecmpt]

We’ve built the fastest inference runtime for LLMs on Apple Silicon, faster than MLX and llama.cpp.

BaseRT is up to 35% faster than Apple's MLX on token generation (decode) for small models, and 78% faster on prefill for bigger models.

Details below.

[Lukas Wesemann @lukasletsgo]

All Macs in the world added up are equivalent to 6% of AI data centre compute on Earth.

Around 120 million active Apple Silicon Macs, call it 8 TFLOPS of FP16 GPU compute each. That is roughly 970,000 H100-equivalents, against 15 million-plus in the world's data centres.

[Lukas Wesemann @lukasletsgo]

@mlaiaus is it a job board called Steve?

[MLAI Aus @mlaiaus]

MLAI Team Meeting 🫡

[Lukas Wesemann @lukasletsgo]

@believeapp @launchacoin @launchacoin lol

[Lukas Wesemann @lukasletsgo]

@mlaiaus needs to be fixed asap

[Lukas Wesemann @lukasletsgo]

@mlaiaus check out https://t.co/QeuhpmP8r6

[Lukas Wesemann @lukasletsgo]

@mlaiaus Let’s goooo https://t.co/vaAxiAoQhL

[MetaOptics @TMOS_ARC]

Finally, congratulations to @LukasLetsGo for winning TMOS Team Member of the Year! Lukas has spearheaded our flagship program, launched a start up, and provided support for the team of students he supervises. Well deserved!

[MetaOptics @TMOS_ARC]

TMOS Flagship Program Manager @LukasLetsGo presented an invited talk at the IEEE conference in Spain this week!

[MLAI Aus @mlaiaus]

Join us for the Green Battery Hackathon 2024 - Get your ticket on https://t.co/lMEPc9GZHi

#Melbourne #Sydney #ai #EnergyTransition #DataScience

[Lukas Wesemann @lukasletsgo]

@igordownunder @ACSPhotonics @SPIEtweets @TMOS_ARC @UTS_Science @UTSEngage Congrats Igor! Well deserved.

[Lukas Wesemann @lukasletsgo]

I think experiments like this one show the potential of AI assisted scientific ideation in the future.

Check out our prototype here and let me know what you think https://t.co/VkS6HWS4MW

[Lukas Wesemann @lukasletsgo]

It shows you the collaboration ideas it generated, and colour codes how each of the two researchers would contribute their skills.

Some of the ideas are actually decent, a lot of them are bad, and some are funny.

[Lukas Wesemann @lukasletsgo]

But I think it can get you thinking, it can help new PhD students experiment with ideation, and it can give you a conversation starter if you just met someone at a conference.

[Lukas Wesemann @lukasletsgo]

@PhillipHaeusler and I decided to code up a little prototype based on openAI’s GPT API. We called it wcwdt (what can we do together).

[Lukas Wesemann @lukasletsgo]

What if we had a tool that we could give two researchers, it would read their Google Scholar profiles, understand concepts for ideas that have proven successful in the past, and generate specific collaboration ideas based on that?

[Lukas Wesemann @lukasletsgo]

💡 Exciting news - we built a prototype for AI enabled science collaboration. “What can we do together?”:

Imagine you just met someone at a conference, and you already had three conversation starters for potential collaboration projects ready to go.
#ai #science #OpenAI

[Lukas Wesemann @lukasletsgo]

@theaiacademic incredible achievement!