I'm tackling part of the issue of food toxin remediation with my new venture, NeutraOat (neutraoat.com). It's a modified oat fiber supplement that selectively traps BPA, PFAS, and plasticizers in the gut and reduces levels in the blood serum.
The funding for this is tough, though. Everyone loves the idea, but it's difficult to find people to fund R&D to make sure the product actually works over brand building and marketing. I've had to be very scrappy. Hopefully this will change in the future as we build momentum and awareness, but for right now it's tooth and nail.
I actually had a battery for a drill meltdown on me earlier this year. If I hadn't been home (and it hadn't been on my stone counter when it happened), I probably wouldn't have a home.
If you're going to use SQLite as an application file format, you should:
1. Enable the secure_delete pragma <https://antonz.org/sqlite-secure-delete/> so that when your user deletes something, the data is actually erased. Otherwise, when a user shares one of your application's files with someone else, the recipient could recover information that the sender thought they had deleted.
2. Enable the options described at <https://www.sqlite.org/security.html#untrusted_sqlite_databa...> under "Untrusted SQLite Database Files" to make it safer to open files from untrusted sources. No one wants to get pwned when they open an email attachment.
3. Be aware that when it comes to handling security vulnerabilities, the SQLite developers consider this use case to be niche ("few real-world applications" open SQLite database files from untrusted sources, they say) and they seem to get annoyed that people run fuzzers against SQLite, even though application file formats should definitely be fuzzed. https://www.sqlite.org/cves.html
They fail to mention any of this on their marketing pages about how you should use SQLite as an application file format.
Breathing and sleeping in negative ionized air for 40 years. Just high voltage potential -7 to -10kVDC and spread carbon hairs (better than metal spikes) to launch it.
Walls and floors always have positive charge conducted from the ground outside relative to the air, no sparks means ozone in too small concentrations to worry about. Dust, smoke, bacteria and viruses stick to walls not the inside of lungs and the air is clean and odorless. You can shine a very bright flashlight through it in a dark room and see absolutely no beam. Every so often you sponge off the walls with strong cleaning solution. Latex paint stains easily near the device which is a subtle way of reminding you how germy it really would have been. Use plastic over walls near the device to save yourself some color matching and painting.
Over all these years, the most annoying thing has been other people trying to sell me HEPA filter solutions with screaming fans that need accessory replacement often. They insist I'm killing myself with ozone as a fear tactic. Few people sell just ionizers or sabotage the concept by selling weak/ineffective ones... because HEPA is big money.
Ionizers use tiny energy and no recurring supplies. Just make sure your electronics are grounded well.
"Conservatism consists of exactly one proposition. There must be in-groups whom the law protects but does not bind, alongside out-groups whom the law binds but does not protect."
You could be the first if you were to develop an eval (preferably automated with llm as judge) and compared local deep research with perplexity's, openai's and deepseek's implementations on high-information questions.
Increasingly hard to get Mg tablets without Vitamin B and so increasingly easy to get excess B12. Australian formulations of magnesium sold by Swisse don't have B. The same branding in HK does. I wrote to them about it, interesting response: "check the therapeutic goods administration site for canonical ingredients we don't commit to labelling the jar" they do list the neutral filler and Mg they just don't say "no B12"
B is typically included for synergistic uptake reasons I believe.
Excess B12 is really bad. Neuropathy. Considering you take Mg to get rid of muscle cramps (pain, conducted by nerves), a bit ironic. It's increasingly common in Australia now.
Excess Mg is unlikely. I think the body chucks it out in Urine pretty rapidly. I still take it, the cramps after sport return within a few months when I stop. I have a fruit and veg rich diet, I get plenty of natural Mg and K which is also good for muscle cramps. Age related insufficiency I suspect (63) malabsorption comes with age.
The solution for me to eliminate headaches when working at computer screens was getting an extra set of intermediate distance glasses specifically for computer work. The "computer screen distance" of 3 ft is in between book-reading distance of 1 feet and driving distance 20'+ feet. I also avoid progressive lenses or high-index lenses for computer work. I commented about how arrived at this solution previously: https://news.ycombinator.com/item?id=15375221
Reading glasses work fine when the screen is very close to your face such as a laptop screen. However if it's a separate monitor that's ~30 inches away, reading glasses are slightly blurry which can lead to eyestrain and headaches.
I wish I could draw like that! The artist is Hanja Grgičević and she works for a sister dev company of the publisher Raw Fury. I gave Hanja my mood board (https://www.canva.com/design/DAF1ZIStuTY/j66Hl6xPyMNU6cfRsw2...) and we iterated from there. I'm so happy with her work :)
It's fun to host at home, I run docker on alpine VMs on two proxmox machines. Yeah, different docker machines for each user or use-case look complicated but it works fine and I can mount nfs or samba mounts as needed. Only thing I have on the cloud is a small hetzner server which I mostly use as a nginx proxy and iptables is great for that minecraft VM.
Why did you go for Cloudfare tunnel instead of wireguard?
Just get modules or strip from Yuju; I have zero bad experience over many years with them.
Sure, it's not cheap, but IMO ~1$/W plus simple PSU (either some standard voltage, or dimmable current, depending on what you got) for good and 1.5~2$/W for very good CRI is quite fine.
Not quite. I've been taking Beta-hydroxybutyrate (BHB) salts in powder form, mixed with water. Here is some interesting literature discussing its benefits as a metabolic regulator of proteostasis in aged and Alzheimer-diseased brains:
Lifting weights can certainly help, though people who focus on the bench press and neglect their upper back muscles may find that their posture actually gets worse, rather than improves.
(Yet it's interesting, though hardly relevant for most of the population, that bulky upper back muscles can also look like bad posture. In extreme cases, a very well-developed upper back can even result in a mild hunchbacked appearance.)
More to the point, those posture correction devices actually work. The shoulder straps are okay. The Chinese sell a slim metal device -- I've never seen it in the USA -- that fits around your neck and prevents you from looking down and slouching over. That damn thing is like magic. A few two-hour sessions over a few days can markedly improve your posture.
x2 in stereo via optical, but a single 11 knocks an A4 out of the water, they're not kidding when they say carlsson sound, it is just that - OB-4 sounds fine but fell over and broke the dumb spinny volume knob thing 3rd day, back to TE and huge wait... never mind they're literally NOT FIXING KO-2s right now, refunds only, fuckin' TE.
1. Lexical search with BM25 alone gives you very relevant results if you can do some work during ingestion time with an LLM.
2. Embeddings work well only when the size of the query is roughly on the same order of what you're actually storing in the embedding store.
3. Hypothetical answer generation from a query using an LLM, and then using that hypothetical answer to query for embeddings works really well.
So combining all 3 learnings, we landed on a knowledge decomposition and extraction step very similar to yours. But we stick a metaprompter to essentially auto-generate the domain / entity types.
LLMs are naively bad at identifying the correct level of granularity for the decomposed knowledge. One trick we found is to ask the LLM to output a mermaid.js mindmap to hierarchically break down the input into a tree. At the end of that output, ask the LLM to state which level is the appropriate root for a knowledge node.
Then the node is used to generate questions that could be answered from the knowledge contained in this node. We then index the text of these questions and also embed them.
You can directly match the user's query from these questions using purely BM25 and get good outputs. But a hybrid approach works even better, though not by that much.
Not using LLMs are query time also means we can hierarchically walk down the root into deeper and deeper nodes, using the embedding similiarity as a cost function for the traversal.
On my LinkedIn post about this topic someone actually replied with a superior method of steering LLM output compared to anything else I've ever heard of, so I've decided that until I find time to implement their method, I'm not going to worry about things.
tl;dr you put into the prompt all the JSON up until what you want the LLM to say, and you set the stop token to the end token of the current JSON item (so ',' or '}' ']', whatever) and you then your code fills out the rest of the JSON syntax up until another LLM generated value is needed.
I hope that makes sense.
It is super cool, and I am pretty sure there is a way to make a generator that takes in an arbitrary JSON schema and builds a state machine to do the above.
The performance should be super fast on locally hosted models that are using context caching.
Eh I should write this up as a blog post, hope someone else implements it, and if not, just do it myself.
I use the Programming, Forums, PDFs, Recipes, and Small Webs lenses pretty regularly, though I haven't tried making my own lens yet. The 'Programming' lens is probably what you'd want for your Python script.
Note: I work part-time at Kagi (was a Kagi user before that), not doing search stuff.
I've been (semi-)intentionally unemployed for a little over a year now, and I reached a point a couple months ago where I just LOST IT at the prospect of having to continue to make decisions at every moment of every day about what to do next. I even have a daily checklist of about 15 things I need to do every day to keep making at least minimal progress toward my goals, but because none of them had assigned times, I was still having to choose every. damn. thing.
So I decided the list should be done before lunch, which meant I could pick anything on the list, but it had to be something on the list. And after lunch, I have a list of three big-picture projects I can work on. I can choose any one of them, but it has to be one of them. Their next steps are reasonably well laid out, because I don't think I do generally have executive dysfunction issues, so I get to pick one of three projects and work on it until dinner.
This has saved my sanity. I don't waste very much time at all now. I don't do well with strict routines, but these buckets take so much of the guess work out. In the morning, 15 options. In the afternoon, only 3. It's so much more manageable.
I feel this, and not even in a situation that had complex transparency needs. I went down a path that needed the surrounding background perfectly matched to the video background, and it ended up being easier to render the video to a canvas, get the rendered pixel value, and update the surrounding background. Still necessary even knowing what the video background color should be given browser differences in rendering the video’s color. Prompted a fun conversation internally, though[1].
2. Ultimately, reform and speed FDA approval for fatal diseases like recurrent / metastatic head and neck cancers: https://jakeseliger.com/2024/01/29/the-dead-and-dying-at-the.... A drug like petosemtamab (MCLA-158), which I was on from Sept. 27 2023 to March 29 2024, should already be approved, instead of continuing to wander around in clinical trials.
Hacker News, thank you for all the links and all the great reading. Now I have to say goodbye.
I’m with my wife Bess (https://bessstillman.substack.com/) and my brother Sam, and crying, but it is okay. At the end of Lord of the Rings Gandalf says to the hobbits, "Go in peace! I will not say: do not weep; for not all tears are an evil.” And that is how I feel now. Ending prematurely hurts, but all things must end, and my time to end is upon me.
Calming the sympathicus is one of the very few things that somewhat works for me. It helps to relax and get good sleep.
One technique I'll add to the list: skin massage (literally pinching it with your fingers) around the sympathicus area: along the lower ribs, up the sternum and into a V-shape around the neck (roughly following the collarbones). The nerve endings on the skin literally stimulate the sympathetic system, allowing it to "reset". I don't know the exact mechanism, but I can vouch for it actually working.
We know how myopia works now. Low dose atropine and glasses with fogged edges reliably slow or stop myopia progression. Myopia is not reversible by natural means. Sunlight is maybe related but it is more likely the proven mechanism of near work, specifically keeping peripheral vision in focus which tells the eye to grow longer (there is no brain involvement, the eye does this alone). Myopia discussions are terrible, they are filled with people offering advice and guesses on something they don’t know anything about, and people claiming it’s curable.
Thank you :). I updated the README to have some more explanation of the steps.
The chunking algorithm chunks by logical section (intro, abstract, authors, etc.) and also utilizes recursive subdivision chunking (chunk at 512 characters, then 256, then 128...). It is quite naive still but it works OK for now. An improvement would perhaps involve more advanced techniques like knowledge graph precomputation.
Reranking works by instead of embedding each text chunk as a vector and performing cosine similarity nearest neighbor search, you use a Cross-Encoder model that compares two texts and outputs a similarity score. Specifically, I chose Cohere's Reranker that specializes in comparing Query and Answer chunk pairs.
As an aside, starting Vitamin K2 a few months ago has given me a noticeable increase in energy and focus, make of that what you will. Vitamin K2 is one of those somewhat hard to get vitamins in a typical American diet.
My sister had RA in her early twenties. For years no treatment was able to restore a reasonable quality of life. There were many side effects which needed treatment as well. She went from doctor to doctor, while she lost the best parts of her life.
An unexpected path led her to a tropical diseases doctor who had positive results with RA patients by using an unconventional treatment not supported by mainstream medicine at the time: He tested the patients blood for bacteria (by cultivation in a lab) and prescribed an antibiotic course that matched the detected bacteria. This treatment (testing + prescribing antibiotic)was repeated in regular intervals over the course of a whole year.
Her RA flare-ups stopped and recurred only in ever longer intervals which were arrested by this treatment. After a year the progression of the RA had stopped.
This treatment is controversial because it is considered bad practice to prescribe antibiotics liberally, for several good reasons. Mainstream medicine did not have a theoretical foundation to justify this treatment path.
The tropical disease doctor founded his treatment on the empirical research by some research factions that the “auto-immune” theory of RA was not showing the full picture.
The chiropractor I'm talking about asked a lot of questions and gave me a lot of advice about what to do at the gym:
- get a trainer for a while and
- work on stabilizer muscles
- pull 3x as much as pushing
- warm up before the training session so as to get the most of the trainer
- stretches to do after the workout
- etc.
I've never had a doctor of any kind be as helpful as that. All GPs nowadays see you for just 5 minutes. Specialists see you for a bit longer, and even then there's barely enough time to cover half of what the chiropractor I'm talking about did. Now, one of the things about that chiropractor is that she speaks really fast, so there's that.
I totally agree. I use ffmpeg's dynaudnorm[0] filter in mpv/IINA when watching media to better normalize sound volume. It's the best audio normalizer I’ve used. No loss of dynamic range within the target window duration. (~15 seconds by default, but I tweak it to around ~8 seconds)
I'm not so sure, my experience tells me differently..
I am a thin guy with very low body fat. As soon as I told my doctor that I had woken up 2-3 times with my soft pallet dry & semi-collapsed, also that I would wake up regularly with a dry mouth, he immediately sent me for a sleep test. I went and did the test and the results were that I had mild sleep apnea. The sleep doctor told me that they were ordering a CPAP machine for me. I asked why I needed it for mild apnea and also told them I wouldn't use it. They ordered it anyway and told me to take the issue up with my doctor.
While all this was going on I was in the process of reducing the amount of meds I took (bad accident in 2010). I also started reading up on & studying sleep apnea. One of the first things I learned, and that worked, was keeping my mouth shut during sleep (many ways to accomplish that). The other thing I learned is that Big Pharma/FDA does a horrible job of listing medication side effects. I say that because "causes sleep apnea" was not on the list of side effects for carbamazeine (used for my peripheral neuropathy). As soon as I stopped the carbamazepine the apnea symptoms vanished. I have no need for a CPAP, it was ordered and I've received many calls about it, but there it sits at the sleep doctor's office..
The funding for this is tough, though. Everyone loves the idea, but it's difficult to find people to fund R&D to make sure the product actually works over brand building and marketing. I've had to be very scrappy. Hopefully this will change in the future as we build momentum and awareness, but for right now it's tooth and nail.