PLBjt

Is the CEO technical or not?

Hi, depends on what kind of model you want to fine-tune. Could you be a little more specific?

Hi HN! I'm PLB from phospho. I’m building a dev kit for ML engineers to work with real-world robots using bearable dev tools (Python, HTTP APIs, WebSocket…)

Some Python examples: basic robot controls (https://docs.phospho.ai/examples/control), rock paper scissors against the robot (https://docs.phospho.ai/examples/vision#rock-paper-scissors), Huggingface AI models (https://docs.phospho.ai/examples/ai-models).

The kit bundles a SO-100 robotic arm, a Raspberry Pi as the server and control module, a stereo camera, all the wiring. The control module handles all the low level stuff: PID control, inverse kinematics, trajectory, calibration, torque limits… You also get a Meta Quest app to remote control the robot. This brings down the time to get started from 100+ hours to 2 minutes.

My team is making a small batch available at $495 each for early builders who want to experiment. You can buy one on our website (refundable until shipped).

I’d love your thoughts, experiences, and critiques. If you have ideas on what you’d use our dev kit for, or feedback on how to make these systems more accessible, I’m hanging around in the comments for a few hours. Let me know what you think!

Key highlights: - Uses a simple linear transformation on existing embeddings - Boosts hit rate from 89% to 95% on real-world examples - Minor increase on latency (less than 10ms) - Works on top of blackbox embedding models (Mistral AI, OpenAI, Cohere,...) - No dataset needed (just your documents) - Train easily on CPU

The safety filter of GPT 4 vision and Claude 3 Opus is critical for user facing products, but to build internal tools it impacts a lot performance (eg. to build an image moderation LLM). We had this problem internally and we had to build our own Llama 3 8B with vision (Llava architecture). We are now releasing it as an API for other developers experiencing the same issue with too censored multimodal LLM APIs.

Portkey is a proxy, so you will have to call their API, and they will route the request to a LLM provider. Then, they provide you with some observability tools to look at these requests -> LLM router + observability phospho is not a proxy, you log the messages and feedbacks to our API. We then extract insights and detect events inside the text -> text and product analytics