vermaat

Hetzner S3 Coming soon.

"We charge a base price of €0.0096/hour for every hour of Object Storage runtime, up to a maximum of €5.95/month. You receive 1 TB of storage and 1.5 GB of traffic included per hour of Object Storage runtime. If the included quotas are exceeded, additional storage is charged at €0.0096 per TB hour and additional traffic at €1.19 per TB."

This includes 19% VAT.

Cloudflare R2 is priced differently. It will be hard to beat R2, but at least the Hetzner option is nice for users who want to storage everything within the EU / close to their Hetzner servers.

Nice! Looks similarly written like a LLM prompt to carefully instruct people. It even includes the tipping haha

Wish I could also use this as external MacBook usb-c monitor so I can code during the day.

Great that it drastically reduces costs vs GPT4! Curious how the LLM model (custom Phi-3) is at extracting knowledge entities when it is dealing with not every day knowledge. The smaller Phi-3 models are good at reasoning, but probably lack some specific domain knowledge that e.g. GPT4 has? So this would make connecting graph entities harder?

Nice work. Any details how it compares to Redis in terms of performance etc? (besides avoiding an extra service when you already have Postgres running)

Cool! I was also working on a prototype that uses text-based reinforcement learning techniques. This has a different use-case, but good to read anyway.

Looks great

Maybe not exactly what you are looking for but BestTime provides relative foot traffic data (also for Montreal's subway station:

Here a free demo page of some subway stations in Montreal:

https://besttime.app/demo/radar?q=metro+stations+in++Montrea...

Interesting question. While higher water (due to global warming) is a threat to the Netherlands, it feels like The Netherlands may be more ready for it than other countries. The Netherlands has already been dealing with high water for centuries, and as another comment mentioned: half of the country is already below sea level.

The intesting part: The flood defense closes automatically at a water level of 3 meters above NAP (Amsterdam water level baseline) at Rotterdam or 2.9 meters above NAP at Dordrecht. The two doors, each 210 meters wide and 22 meters high, fill with water when the barrier closes. They then sink to the bottom in about two hours. A computer system automatically determines whether the barrier should be closed. However, a team from Rikswaterstaat is present to supervise.

Noob here. How is this different than reinforcement learning libraries like: OpenAI’s Gym TensorFlow’s TF-Agents ReAgent by Meta DeepMind’s OpenSpiel Amazon SageMaker RL

I hope the 13,5 million Euro funds will be used wisely for the GPT-NL project. It would be nice if they open-source it (or at least the training dataset).

Works great, does indeed have more Dutch knowledge besides only speaking Dutch.

How do you compare your own trained LLM versus using for example GPT4 + RAG (Vector DB + your Australian Law DB?

Nice. Maybe add Speech-to-text capabilities so the user can use the browser using their voice (maybe while holding a key)

Very cool. Good luck!

Nice work. I'm trying to build something similar, yet different: FlowFlow.AI: Automate your core business workflows by defining step-by-step tasks and follow up actions for the AI agent to execute. - The AI agent decides the appropriate next step in the workflow based on one of the predefined conditions. - Keeps the AI in control within the boundaries of your complex workflows. - Use the FlowFlow API or Python/JS packages to interact with the workflows in your code. FlowFlow.AI is almost ready for the beta launch. :-)

Nice work. I'm trying to build something similar, yet different: FlowFlow.AI: Automate your core business workflows by defining step-by-step tasks and follow up actions for the AI agent to execute. - The AI agent decides the appropriate next step in the workflow based on one of the predefined conditions. - Keeps the AI in control within the boundaries of your complex workflows. FlowFlow.AI is almost ready for the beta launch. :-)

I tried the 'Circadian lighting component' before on my Home Assistant setup. However, the component made my lose control over certain devices (even though I configured it that I should be able to override the temp/brightness settings). It either kept lights off when I turned them manually on, or visa versa. So now I just reverted back to 'Automations' per room that are triggered by motion sensors and activate one of the predefined scenes. Each room has aprox 3 scenes: day, after sunset, and night. The automation trigger the scenes based on predefined times. I agree with the author that it's really great to not get blinded by bathroom lights in the middle of the night. The automated motion activated scenes per room/ time also give a very luxurious effect when e.g. entering the living room in the evening and it automatically toggles a cozy dimmed light scene.

Same for me. Funny enough the Cloudflare's "Always Online" TM functionality seems to work better for my app's website than for Cloudflare's website. It shows my app website, but the CSS is not loaded....