HN user
Wow this is badly vibe coded. Check your submission on a couple of devices (desktop, mobile) and browsers.
How did you implement the tracking feature? Do you plan to narrow down in one legal sector where this could be especially useful?
Radiologists very often have to weigh up different theories, guidelines based on the symptoms. The certainty of their diagnosis is their added value, or if they don’t know they will tell you why.
An AI telling you it could be X or Y because theory ABC… is the academic answer and a luxury clinicians don’t have. AI doesn’t give you what you want. I don’t see any added value in using generic AI models for this
Exactly, try can get a very limited FOV which is probably why they showcased it on arms/legs first
You’re completely right, this is why currently ultrasound reconstruction happens on FPGAs. They would need a lot of them given the number of transducers. https://pmc.ncbi.nlm.nih.gov/articles/PMC6057541/
MRI physicist here as well. I have a basic understanding of ultrasound, and this looks like an array of transducers organized to perform tomography, just as CT did for Xray.
However Ultrasound quality depends highly on transducer-skin contact.
Any physicists here to comment on the effects of sonar through liquid and the effects on image resolution and field of view?
This has factored out product development, which is more than compute resources. Just like any industry, some organisation needs to take ownership and responsibility to convert technology to a usable product.
That only tells what base architecture they used, but fine tuning does not increase the number of weights, it just adapts the weights to improve better on a fine tuning dataset- something they claimed they had done
Nice tool. I like the local approach. I think a nice feature would be to remove all PII from documents, so that users can redact PDFs and upload to their favourite LLM.
They used a custom neural net with autoencoders, which contain convolutional layers. They trained it on previous experiment data.
https://arxiv.org/html/2411.19506v1
Why is it so hard to elaborate what AI algorithm / technique they integrate? Would have made this article much better
The pain already starts when a new feature needs to be introduced, your colleague is assigned to the task and the architecture is completely unfit for modular development.
Any sensible experienced programmer will take into account possible future features spoken about around the office, when deciding how implement a feature. To me that’s the key reason I can’t delegate AI complex features that will grow. The same reason I inspect and clean the toilet when I’m done - out of respect to my colleagues
Great write up and recognisable performance. For a pipeline with many (~50) build dependencies unfortunately switching interpreter or experimenting with free threading is not an easy route as long as packages are not available (which is completely understandable).
I’m not one of these rewrite in Rust types, but some isolated jobs are just so well sorted for full control system programming that the rust delegation is worth the investment imo.
Another part worth investigating for IO bound pipelines is different multiprocessing techniques. We recently got a boost from using ThreadPoolExecutor over standard multiprocessing, and careful profiling to identify which tasks are left hanging and best allocated its own worker. The price you pay though is shared memory, so no thread safety, which only works if your pipeline can be staggered
I still use a high school calculator for doing pen and paper maths. Stuff like figuring out when functions change sign. I find it less distracting than a code editor