Welcoming two advisors to Piris Labs:
Ali Alasti, Oracle SVP, 24 yrs building datacenter-scale hardware.
Marc Baldo, Director of MIT RLE, pioneer of high-efficiency OLEDs.
The rise of open-weight models in engineering design!
Cadence is using GLM for circuit design, and weโre seeing interest in laser and EM simulations.
Excited to see our photonics infrastructure help design the next generation of chips!
Building similar tools? Letโs talk!
We are in a Y Combinator startup event in Taipei! ๐น๐ผ
Seems like a great event, with tours of major players in the semiconductor ecosystem.
If youโre around and would like to meet, let us know!
#YCombinator#YC#Taipei#Taiwan#Startups#Founders#TechStartups
We tried Pika on our AI infrastructure!
Building similar AI products and looking for a partner to run them at the speed of light using photonics? Weโd love to work with you!
Explore our AI endpoint: https://t.co/Re0QUIC3Cb
Building unique AI solutions? @PirisLabs could be your partner!
Weโre running a large proof-of-concept AI cluster with full-stack memory and photonic interconnect solutions, from chip to software.
Try our AI endpoint: https://t.co/7BovRM19Pm
Need a different experience? Looking to reduce your cost per AI task? Reach out!
Partner with us: [email protected]
๐ Weโre hiring at @PirisLabs!
Member of Technical Staff โ Photonic Packaging & Laser Scientist.
๐ฐ $200Kโ$500K + equity.
Come build the future of photonics and AI infrastructure with us!
#Hiring#Photonics#Lasers#AI
Open-weight models are interesting because they allow us to co-optimize hardware and software, experiment with new architectures, and push the boundaries of intelligence per watt. @PirisLabs
Marvellโs massive acquisition of @Lumentum is another huge validation of what we're building at @PirisLabs.
As optical transceivers and interconnects demand lower power and higher bandwidth, the industry is increasingly moving in our direction
@TechCrunch The founders' clarification is the interesting bit: Nvidia for prefill, specialized silicon for decode. Then moving KV state becomes part of the economics. The test is end-to-end cost at a fixed latency target under real traffic, including transfers and idle capacity.
@ekuyda@IVP@a16z@ForerunnerVC@Benioff@tobi@dickc@marissamayer Shared AI makes memory much more interesting: a preference needs an owner, a source, and an expiry. 'We usually do Fridays' and 'Friday is confirmed' cannot become the same fact. Keeping that distinction across people and sessions feels central to making this dependable.
@morqon@CheekyPint Pooling only wins if the memory is shareable. A cloud fleet can have spare compute while long-lived KV caches fill its HBM. For startups, the useful comparison is cost per completed task under privacy and latency constraints, with cache residency and data movement included.
@snowmaker Threads also give agents a natural unit of persistent state. The tricky part is separating conversation history from current task truth: what was decided, what changed, and who owns the next action. Otherwise a faster agent just re-reads a longer thread on every wake-up.
@AravSrinivas Knowing when to abandon a plausible plan after contradictory tool results. More tokens can deepen the wrong path. A useful test: hold context and tools fixed, increase the reasoning budget, and see whether recovery improves. That separates reasoning limits from serving limits.
A little while ago, we at @PirisLabs started using GLM5.3 through our own endpoint for a few simple tasks. We all were amazed by how much better it felt (subjectively at least) and how much faster it was than the latest models from ChatGPT and Anthropic. On top of that, it costs less than a tenth of the price. Today, We are using our own endpoint for everything, from coding to task completion, with almost no reliance on any of the big labs' endpoints at all. Strange feeling! if you like to try it yourself you can go here: https://t.co/7BovRM19Pm