HN user
I did use default settings on those tools. I'd agree it wasn't the best benchmark and certainly wouldn't hold up if the goal here was to break any records in compression. The tools and settings chosen were mostly to validate 'this ended up roughly in the same ballpark as other tools' over 'Claude burned a lot of tokens and still made a terrible compression tool'.
One day I hope to do enough that Google doesn’t suggest my name is a typo.
The Gnorp Apologue (mentioned in another thread here) was also notably written in Rust. https://store.steampowered.com/news/app/1473350?emclan=10358...
A very exciting release and I hope it stacks up in the field. I ran into their team a few times in a previous role and they were always extremely robust in their clinical validation which is often lacking in the space.
I still see somewhat of a product gap in this whole area when selling into clinics but that can likely be solved with time.
I'd recommend giving Raylib a look if you haven't. I've been working on a couple small ideas using it lately and it feels very aligned to being a game 'engine' for people who come from a more classic software engineering background.
There was a time where I felt like this. Especially early as a founder as we began to bring on staff for the first time. I started to feel that if I wasn't building the things then someone would call me out for not being deserving of the founder title any more.
It took a long time to grow out of this feeling. These days I get a lot of joy from management. I try to look at myself like a coach on a sports team. I may not be on the field but I feel good when we win knowing the work I have done to help my team succeed.
I would agree with the author that the feedback loop is different. It is also often slower and more subtle. That doesn't however mean that it's not there. Perhaps the author is new to management, perhaps the author just really likes being an IC. Their journey is their own but I would certainly say there are many paths to endorphins as a manager, even as a manager of managers.
This looks good as an overarching framework but will likely fall into the same bucket as a lot of regulation in cutting edge fields (if this ever becomes mandatory). These are only as good as the quality of the people doing the assessment.
I've worked with a lot of people in the medical world while developing SaMD (software as a medical device) in the past that had little to no idea about software. They can apply the principles in the abstract but will likely not dig deep enough to catch some very major issues.
In the medical world, things like post market surveillance and notification of adverse events help to at least create a public feedback loop here. I think we will need something similar in this space if we really want to see more than a surface level, checklist ticking exercise.
High Output AI - https://www.youtube.com/channel/UCzuyWHAoUBXJBfVlDRJzbIA
A (mostly) weekly podcast we've been doing for almost a year now. Trying to cover a mix of next, papers and releases through the lens of two previous founders of an AI company.
There is a talk about that issue [1].
During my PhD this issue came up amongst those in the group looking into compressed sensing in MRI. Many reconstruction methods (AI being a modern variant) work well because a best guess is visually plausible. These kinds of methods fall apart when visually plausible and "true" are different in a meaningful way. The simplest examples here being the numbers in scanned documents, or in the MRI case, areas of the brain where "normal brain tissue" was on average more plausible than "tumor".
[1] http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_...
To Cal's credit, he has somewhat done this. He has a playlist on YouTube [1] which covers 90% of the lessons and it's pretty short. There is a lot more material sure but it's mainly focused on smoothing out that last 10% and can easily be skipped.
[1] https://www.youtube.com/playlist?list=PL8xK8kBHHUX43VVxO3b7s...
Something I will add which relates to the examples in the post (gym, business, relationships) is that big things happen with lots of small steps over time. You can't do a years worth of workouts on Jan 1 and be fit for the year. These things take daily effort and cycles of work and recovery to happen.
That desire for perfection can also be a desire to be done. To have it finished and get closure. It's hard to accept that some things are going to take a long time or a lifetime.
A model I've uses is one of risk. Whatever your idea is there are going to be a number of potential reasons it might not work out.
For some ideas the biggest risk is the market, does anyone want this? For others it might be the technology, is this even possible?
In a lot of cases you can do things to reduce those risks and give yourself more confidence to jump ship and go all in. That might be talking to people, it might be reading academic papers. In some cases, but honestly pretty rarely, it will be building something. Most of the time that can come later.
Not dismissing any of the great advice in here already around non-compete, company IP etc. most of this kind of work (research, conversations etc) isn't something your employer can "lock down" as concretely as an MVP. Plus if it is tangential to your current job and your idea doesn't work it, you may still end up better off from having learned something new.
The wiki is fantastic, I think I landed on the gitbook version somehow earlier this year. Love seeing long term projects like this full of interesting content. Well done.
The post on Research and Adventure is great. It reminds me a lot of my time at Grad School. Some beautiful photos as well.
The content and the design of your site look excellent, well done. I think the lab notes format for longer term projects works well to group ideas.
Is there any way to preview what is in the member magazine? It looks compelling but hard to tell if it crosses into purchase territory from the outside.
Writing mainly about AI and healthcare these days. Using ghost (hosted) after many years of picking a new platform every time I felt compelled to write more.
Yeah, very reasonable question. I suppose a somewhat simple version could be to add exercises to a 'cart' and then change the colour muscle groups on the home page based on cart contents. Would then be pretty easy to click back and forth and colour in the desired groups. If I could then look at my cart when I'm happy I have good coverage that would probably be enough for my needs of 'does my workout plan have any blind spots'.
Great website and kudos for making it. Noting you have limited spare time feel free to ignore but: I would love to be able to select a collection of exercises and see a visual of which muscle groups I do and don't have covered in the chosen exercises. Perhaps with some suggestions on exercises to cover those that are missing.
I have this bookmarked alongside a couple of other short stories that you might enjoy if you liked this:
- http://www.multivax.com/last_question.html - http://www.galactanet.com/oneoff/theegg_mod.html - https://www.tor.com/2010/08/05/divided-by-infinity/
The process reminds me of 'Ultralearning' by Scott Young, do you have any book/resource suggestions for tackling similar projects?
certainly, diffusion imaging is common in analysis of many physiologies including breast and prostate. It forms one of the sequences in multiparametric MRI which has had some recent success in improving the diagnosis of prostate cancer [1].
I'm also aware of at least one upcoming trial of full (or mostly full) body diffusion imaging to look for sites of metastasis. The trial is still in planning so unfortunately no link to share for that one.
[1] http://thelancet.com/journals/lancet/article/PIIS0140-6736(1...
Not sure which bank you are with but mine seems to have some exclusivity deal with samsung phones which it hides under the guise of 'other phones are not secure enough to do contactless payments'. Not a hard fact but it would also lead me to believe they get some sort of kick back.
Here is one using open CV in python [1]. While there are many ways to find circles the Hough transform is a fairly simple and popular one, I don't know what Matlab uses internally but it would probably be an optimised version of Hough for circles. you can see some more here [2], implementing this process in raw python (with PIL or pillow) was something we did in our image processing course in undergrad so certainly not out of reach if you wanted to give it a go.
[1] http://opencv-python-tutroals.readthedocs.io/en/latest/py_tu... [2]: https://en.wikipedia.org/wiki/Circle_Hough_Transform
Perhaps not its intended purpose but to me it looks like an upgraded version of Google Keep. As someone who likes to collect resources and notes on topics for both research papers and personal projects I see this as a useful tool for that. I don't imagine any of my 'spaces' having more than one or two other people in them.
In reference to all the 'Google will kill this soon' posts. I agree there seems to be a growing list of discontinued products but their Takeout [1] service almost always lets you get your content out. While things like social interactions on G+ might hold value on the platform itself, as long as I have my notes/photos/data at the end of the day, I am happy to treat these products like I would a text editing app, just a tool to work with the data.
You're welcome, not often that MRI related posts come up so it is nice to share this side of what I do.
I think certainly there is room for it and I know there have been some recent entrants into the field which consider themselves MRI R&D. Up to this point though they are mostly bring quite secretive about what they are working on. I think though with the right funding and some smart cash behind a team that knows the space there is room to make big changes and drive price down competitively.
One example of a team in this space is Magnetica [1] who are working on smaller (physically) MRI systems for extremity imaging. (I know them through the university and my supervisor is CTO but I have no personal stake in the company otherwise)
There is 'the' MRI conference on this week in Singapore (ISMRM for those interested) which may result in some MRI related press in the coming days. Not sure where to look for it though as unfortunately the conference content is member only at this point.
Yeah essentially, I meant that in the sense of innovating how people access MRI scans and the data associated with them.
MRI like a lot of other med tech is very heavily patented. The core tech has long since expired but any of the cutting edge work will probably have patents around it. I do believe that an MRI startup has a chance to disrupt but it may be more of an innovation in application, or accessibility than in pure technology.
That's the hope. A little outside my field but current compressed sensing in MRI research is looking at reduction factors of about 2-8 (typically 2, 4, or 8 from tradition) in compressed sensing. This is still pretty rare in commercial machines though.
Some sequences such as EPI [1] can get images very fast (200ms) with a trade off in spatial resolution. This is for things like cardiac imaging where low resolution images of the heard every 200ms are more useful than a single high resolution image. Also for moving parts of the body there is often blurring artefacts due to movement within the image window using longer sequence types.
Unfortunately I think the biggest inhibitor of MRI being more common is still the price of a scan. Here in Australia it can be a few hundred dollars for a scan (and I understand it can be far more elsewhere). Again though if we had access to a cheaper option the idea of regular scans may not be too far off.
I like to imagine a point in the future where the sci-fi full body diagnostic scanning is a reality. Still a lot of work to get there but it is certainly getting closer.
There are a number of machines that don't use superconducting magnets and hence don't have such stringent cooling needs. These machines typically need these sort of cooling setups to support higher fields but imaging is still possible at low field. Some years ago, and I'm by no means up to date with this, there was a group working on imaging using the earth's magnetic field as the main field [1]. While this is the extreme MRI has been functional longer than superconducting magnets have been feasible.
On the flip side of this things like compressed sensing and MRI fingerprinting [2] reduce further the need for a traditional 'full' scan. We are still some way off seeing these sorts of things in clinical MRI but that may be more attributable to the pace at which companies like GE and Phillips move in this space. That and the sheer cost of R&D.
Overall it is a pretty exciting time to be in MRI, despite the outside view being that it has changed very little in recent decades.
disclosure: PhD candidate in electrical engineering, focused on MRI
[1] http://www.ncbi.nlm.nih.gov/pubmed/16828566 [2] http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4376817/
Putting the 'should we or shouldn't we' aside I'd just like to say thanks for putting together an article that goes all the way from a 'classic' setup to something running on a host.
I often find that posts like this are a bit too meta and in trying to write things in a general way they leave out some critical step which makes replicating their ideas difficult.