axg11

44 ms·

511.
by axg11 5y ago
I hope this article ages badly. We're at the tail end of the "mainframe" age of satellites. Expensive, long development times, low-risk projects. Hopefully with launch prices dropping, satellite development can become cheaper
512.
by axg11 5y ago
Perhaps analyze for the combination of dogs + poop? The dog probability will be transient/a spike, whereas the poop probability will be constant (after appearance of the dog).
513.
by axg11 5y ago
Pre-trained CLIP gets you 95% of the way there, so you're correct, fine-tuning isn't necessary to test the market. The one downfall of pre-trained CLIP is that it hasn't been trained on still images from videos. These have a
514.
by axg11 5y ago
Some potential markets: - home security - searching through long home videos - production companies with large video archives (this would require more tooling) I am unsure whether to focus on one of these groups or to go for a more generic
515.
by axg11 5y ago
For sure, here's a few tidbits: ML: fine-tuned CLIP model. Each video frame is embedded using CLIP and then the image embedding is compared against the text query embedding Architecture: everything is serverless using AWS lambda. Basic
516.
by axg11 5y ago
Use the code HNDISCOUNT to get the first month free (100% off). Feel free to cancel anytime during the first month, you won't be charged.
517.
by axg11 5y ago
Did you use an API from one of the big cloud providers? Tingy is based on a custom variant of CLIP, which is quite computationally expensive but worth it for the general querying ability.
518.
by axg11 5y ago
In this example you could query for "poop" and then look at the resulting plot for a sustained period of time with high poop probability? :)
519.
by axg11 5y ago
Interesting point, any idea at what level you would price this?
520.
by axg11 5y ago
These are all great iterative features, but no more than that.
521.
by axg11 5y ago
What would you improve about the iPhone or iPad? I think these products are just mature and there is very little to improve other than iterative updates. The true test of whether Apple has entered their Ballmer era will be if they fail to l
522.
by axg11 5y ago
Ah I didn’t realise that was the rule - I will work on adding a free trial
523.
by axg11 5y ago
Great! Please email support@tingy.video
524.
Show HN: Tingy – search videos with natural language queries(tingy.video)
21 points by axg11 5y ago|22 comments
525.
by axg11 5y ago
First off, I want to make the disclaimer that this project is very much an MVP and the landing page/front end is quite ugly. I'm sharing tingy - this is a service that allows you to upload a video and query it using a text descrip
526.
by axg11 5y ago
Indeed: >You may not tamper with the Glasses, or otherwise obscure or modify any of the features on the Glasses that signal to others that the Glasses are recording (including the external-facing LED light). https://www.facebo
527.
by axg11 5y ago
Does anyone have information on whether the platform will be open for development?
528.
Scientists are creating mirrored biological machinery(arsham.substack.com)
1 points by axg11 5y ago|0 comments
529.
by axg11 5y ago
The free codes have been used up unfortunately. I'll add more when the current queue length becomes shorter.
530.
by axg11 5y ago
Apologies - this should be fixed now. If you have any issues email: support@makeai.art
531.
by axg11 5y ago
Here's my result: https://s3.amazonaws.com/ai.protogenes/art/3ed8a7fa-f60b-11e... "a monkey riding a rocket" I think it works well :) Which architecture is this based on? Did you consider any trips
532.
by axg11 5y ago
Great, I like the name! I'm trying it now.
533.
by axg11 5y ago
This is an interesting idea, essentially a platform to allow people to produce interesting/creative images and distribute from there. I agree with you that this is not a giant leap or value-add from what is out there. I'm hoping t
534.
by axg11 5y ago
That's a great point regarding a free watermarked trial. I'll look into adding that.
535.
by axg11 5y ago
Huggingface is really an incredible org. I hope they continue on the same excellent path. They recently raised $40M and I imagine there will eventually be stronger pressure to monetize their work.
536.
by axg11 5y ago
It's surprisingly expensive to compute, not quite $1 but after stripe fees it's not far off. I'm looking at ways to bring the compute cost down.
537.
by axg11 5y ago
It's a very high-level architecture to that, yes :) I did some fine-tuning on scraped images of modern art to hopefully result in more visually pleasing images.
538.
by axg11 5y ago
For those curious to try but don't want to pay, use the code: MACHINELEARNINGART to try for free (valid for 5 uses).
539.
Show HN: Makeai.art – Create Art from a Description(makeai.art)
19 points by axg11 5y ago|28 comments
540.
by axg11 5y ago
Grail will have to end-to-end validate any changes to Galleri, especially any change of sequencer. Given that any Illumina competitors based on the same IP wouldn't hit the market until 2024 (extremely optimistic) I'd be surprised

More ›