Yep, self-determination is a really really big deal.
But figuring out how to incentivize people to self-determine their way into doing 100% of the things an organization needs to do is a real trick! Some work that still needs to be done just isn't great.
The typical approach is to use a combination of money and "skin in the game" (ownership) to cover those cases. And I think that's essentially the right approach, but this simply won't work for everyone.
And the thing is that it's super tough to separate out less fulfilling work into a distinct job role that can be more highly compensated and done by people who find that more motivating.
I think the best I've seen it work is for managers, who people already expect to be more financially compensated, to fill in these kinds of cracks. If there's something their team needs to do that just isn't getting done without them directing one of their reports to do it, can they do it themselves instead? (And maybe while they're doing it, they can be thinking about whether there is a way someone could build a tool to make it trivial to do the next time it comes up, and then they can probably get their team more excited about building that tool.)
> "caught the improving mentality, or attitude — the one thing all inventors, both then and now, have in common — which had him viewing everything around him in terms of its capacity for betterment."
As much as I resonate with this sentiment personally, I've recently been reading the works of Byung-Chul Han, specifically "The Burnout Society", which has made me really think about how this mentality can be pretty insidious.
EDIT: not that I think this detracts from the thesis of the article itself, I just find it interesting how the author of tfa bolstered their thesis with this right away.
> The concept/idea is not what is patented. The patent is (or should be) for the specific execution of the idea. Competitors are free to implement their feature using methods other that what is covered by the patent, even if the end result gives the exact same functionality.
IP lawyer here (EDIT: not yours, of course): That's a considerable (and potentially-dangerous) oversimplification. What matters is whether what you do comes within the claims of the patent.
I always assumed it was for exactly the reasons you list. Anyone with academic credentials can submit to arxiv, but that leaves out a huge group of non-academic hobby scientists who haven't had a spare 6 years to find their way into the ivory tower.
I may be misunderstanding, but isn’t the technique you describe—using general classifiers to interdict new CSAM and adult content on the child’s device, with reporting to parents—already widely deployed on iOS and largely uncontroversial, since parents can easily adjudicate false positives and edge cases? [0]
I thought the technologies at issue involved either breaking encryption to allow authorities direct visibility into all digital communications, or various flavors of centrally-maintained blacklists of known images, with the idea that the device or the service automatically vets all new user data against the blacklist (e.g. [1]). All of which is impossibly tempting to extend to other types of content that the powers that be find antisocial, inconvenient, or embarrassing.
For what it’s worth, in the US at least, there are jurisdictions that prosecute children as child pornographers for sending images of themselves to their romantic partners (e.g. [2]). I can see how, if I were a parent in that situation, I’d want to know about and deal with that behavior without the fear that my child might spend the rest of their adolescence in prison and the rest of their lives on a pedophile blacklist.
Sad to see this article reach the end of the scrollbar without even a mention of the triazine sunscreens which are pervasive in Japan and approved in the EU with no action from the FDA.
> ADHD is sometimes wittily described as time dyslexia
As someone with ADHD, this wildly misrepresents the actual disorder. ADHD itself has nothing to do with time, it has to do with executive function. Someone with ADHD can try to will themselves into doing something and simply not be able to. They'll try to make a command, and their body or brain will not listen to them. It will completely refuse to do what they tell it to.
This isn't supposed to happen; you're supposed to have control over what you do. You are supposed to be able to decide to do something, get up and simply do it. But with ADHD, it's not that simple. Even things that require no physical action are difficult, because it's not the actual movement that's hard, it's the decision-making itself. Hence "executive dysfunction".
Any "time dyslexia" effect, wrt scheduling and deadlines and etcetara, is just a symptom of it. The reason why people with ADHD procrastinate is not:
- because they don't know what time it is.
- because they don't know when their deadline is.
- because they don't know how much time they have.
- because they don't know how much time they need.
- because they don't know how easy or hard the task is.
It is because their brain wants to do something else more, and it's not urgent yet.
They absolutely cannot work on the task no matter how hard they try. They have not forgotten. They are not slacking off. They literally just can't do it. Their brain refuses to think about it, their body refuses to move for it. They don't have the willpower or the motivation for it. They are trapped. They are completely unable to make any progress because their brain will not let them.
That's what ADHD is.
Not everyone has it this bad, but ADHD is typically characterized by this happening for at least some things. It could be "showering more than once a week", it could be "doing the dishes before 20 of them have piled up in a big stack", it could be "preparing for a road trip days in advance". It doesn't have to be everything, and it doesn't have to be completely insurmountable, but if you have to have a complicated coping mechanism in order to manage to do something that you otherwise can't just decide to do, that's the disorder.
Word of advice to all those who are chomping at the bit to disrupt pharma with AI.
A pharmaceutical company that heavily leverages computation is called a pharmaceutical company. All modern pharmaceutical companies heavily leverage computational tools - including some powered by deep learning.
Any company building a computational platform to accelerate drug discovery / development is not a pharmaceutical company. They are a software company who wants to sell software to pharmaceutical companies, which is a terrible business to be in.
The product the author is selling already exists for free. All of the big knowledge bases use automated ML tooling for curation and extraction. And the people who run them and QC them are world class experts in their domains. I mean, just take a look at uniprot.[0]
And the only types of pharma companies who would want a big knowledge graph would be the large ones with active programs across multiple therapeutic areas. Most of the small companies / start ups tend to be focused on getting one or two assets to market in a therapeutic domain. And the big companies possibly have their own ML teams doing literature extraction - because a team of 5-10 FTE ML / software engineers is like a rounding error in their R&D budget.
The other thing is that the most valuable knowledge is the stuff that is not in the literature. That’s why we do experiments.
The story is about how Tinder and Bumble are in decline. The superficial model isn’t working anymore.
I think the issues are much more fundamental than you suggest. They’re societal and they’re subcultural within the apps. For one, people are much pickier now than they’ve ever been. On the other hand, the dating apps have this filtration problem: those who successfully form a relationship quit the app, possibly for life.
Unfortunately, it’s not random when people form successful relationships. Some people are just much better at it than others. This is where the filtration problem arises: over time, the concentration of people who aren’t good at forming relationships increases, as these are the folks who stay in the apps the longest. This makes it harder and harder to find a relationship through the apps, and frustration ensues.
- Having the APOE e4 allele (which can be mitigated through diet).
Then, more recent discoveries point to the roles of:
- Sleep: detoxifies our brain.
- Exercise: detoxifies the brain and stimulates neuronal growth.
- Cribiform plate ossification through aging: somehow prevents detoxification of substances in the brain.
- Bacteria that causes gingivitis: found to get to the brain, promote inflammation, and its increased presence is found among biopsied brains for thoss who had Alzheimer's.
- Candida albicans, as this paper describes.
Supplements being studied to somehow benefit Alzheimer's:
- Curcumin: supplements like Longvida potentially help, but at doses much higher than those listed on supplement bottles. Other forms are also being investigated.
- Magnesium threonate: gets more magnesium into the brain. Some MDs testify of what they did for their Alzheimer's patients. Still being studied.
Things that probably don't help:
- Donepezil: still being given to patients, but data now suggests it's not very useful.
That's all I can recall from memory. Very interesting disease.
No, don't even post on smaller networks. Post on your own site, is the idea:
> POSSE is an abbreviation for Publish (on your) Own Site, Syndicate Elsewhere, the practice of posting content on your own site first, then publishing copies or sharing links to third parties (like social media silos) with original post links to provide viewers a path to directly interacting with your content.
DevOps/Software Engineer with 20+ years experience. ~100 applications this year, a good number of interviews, but no offers yet. (2.5 calls this week scheduled)
It seems I either get:
- proto-startups with ~5 people and no concrete revenue stream, or
- 5k-person enterprises who want someone with a very specific skillset
Both are fine, but they're picky... as am I, so I'm doing my own consulting for a bit.
What happened to startups? What happened to the thousands of 50-300 person companies who need tech work done? The other day a headhunter called me... because they were bored! Rather different from last year when they were juggling 4-5 excited companies in front of me.
Given my experience, I'll be giving a "Fast Developer/Startup" class to a number of companies. That'll turn into a day-long workshop that will be very valuable :)
Strategies:
- DON'T TAKE IT PERSONALLY
- find roles on e.g. LinkedIn, but _never_ do "fast apply": go to the company's site and apply directly
- use a "highlight positives/negatives" tool to draw certain words in a web page different colors. By interactively seeing a mass of green (or orange) you can quickly make a YES or NO decision on a role. I adore the Chrome plugin Highlight This -- https://highlightthis.net/
- apply for jobs on a schedule (e.g. Mon-Wed-Fri), don't just struggle for hours at a time, it's soul-sucking.
- do Studying (job-related tech) interspersed with Fun Programming (generative art!) -- have fun!
- take care of your health and family, go outside and take walks
If you like ChatGPT for search, I highly recommend checking out Perplexity, it is a search engine that uses the same model but can provide quite good information with source citations.
If you're going for "none os us is smarter than all of us", the flaw in such group decisions is that they are made openly in the group. The first opinion influences the next and so on, and the louder and most persistent opinion gain more weight in the final decision than they deserve. The biases pile up and the quality of the decision goes down.
In order to take advantage of the knowledge of the many, each individual needs to form their opinion independent of the influence of the others. From there you move on to a structured and mediated discussion (i.e., not an adhoc free for all). Of course, participants can change their minds, but they do so based more careful considerations and far less based on the emotions and biases of a traditional group decisions.
See "The Influencial Mind" by Tali Sharot for more details.
There is a grain of truth in it when they say that the IQ of a team equals the IQ of the most intelligent team member divided by the number of people in the team.
I studied Design of Experiments (DOE) in college -- it was part of the core curriculum in my major. The content feels like it ought to be very useful (it's basically techniques on how to conduct experiments efficiently and reducing the number of runs, thus reducing overall cost).... yet I've never had occasion to ever apply it.
One area where I feel it might be relevant to my work is in hyperparameter tuning, especially when function evals are expensive. Instead of doing grid search (which is a brute-force exhaustive search), fractional factorial designs can search the multidimensional space with way less function evals. Yet I don't do it because it's easier to write code to do brute force and let the computer chug away over night.
In the world of web analytics, fractional factorial lets you run multivariate experiments efficiently rather than a combinatorial combination of A/B tests. From online articles, I'm guessing people do use it there, but A/B tests still dominate.
Fractional factorial is also most effective when the covariates are orthogonal and independent, which is rarely the case. We can get around this by projecting high dimensional covariates into lower dimensional space using PCA, which guarantees orthogonality, but this also seems to not be done so much.
Just wondering if anyone is using fractional factorial designs in real life? (or optimal designs like D-optimal designs)
I'm "secretly" pondering a formula where a group of employees share an assistant rather than a manager.
In stead of a layer of management above the people manager the assistants also share an assistant.
With a small salary comes a rigid job description without free styling.
1-4 times per year you bring in a consultant/freelancer to read the reports (AI generated abstractions) and twiddle the knobs for however long it takes. Say 1-2 weeks with nothing but meetings. It should probably involve a hotel, resort or boat trip.
I believe they mean the lawyer hotlinked [1] their image so that every visitor to the lawyer’s page would result in an image download from their server.
The token requirement is a pain. We settled on using Azure Key Vault and AzureSignTool [1]. It costs $5 a month for a HSM key and you can sign things from anywhere.
This is one thing I love about Magit for Emacs. The UI is really clever and slick—maybe the best Git frontend that I've ever used—but the way you interact with UI is by toggling flags and options that actually map to the underlying command line Git arguments. I can seamlessly hop into the command line and feel right at home using it directly.
There's one huge not-immediately-obvious element there: Vercel bandwidth is expensive. My company switched our CMS content over to Prismic in part because serving all media content directly from them was significantly cheaper than doing so through Vercel, even at a fairly low volume.
for something a bit more robust, check out DuckDB. It's a library you can embed, use it to run SQL on local files as well as connect to databases, do joins, analytics, etc.
Many webservers allow you to serve a compressed file (stored on disk) and _decompress_ when a client specifically can't support the compressed encoding. Since most clients should support compression, this means you only use the CPU for the rarer case where the uncompressed data is required.
> Theres a YC video that goes over tar pit problems. If i am not mistaken, this exact scenario is covered as a tar pit.
Yeah the final end goal was either to make lots of deals with local businesses (hard to scale) or to license an ML personality matching model to companies.
The app had a handful of personality questions that I copied over from research done at one of the Nordic universities (I forget which one) on what makes people get along together in a casual setting. The American universities have mostly done research on group cohesion in corporate settings, which maybe hints at what is wrong with American society at large!
Cruise liners and casinos would pay a fortune to know what guests would vibe together.
I had a partner website that allowed for self onboarding, but of course b2b2c is never that simple. :)
The operating costs were so absurdly low (~$200 a month per city it was running in) that letting people create their own events for free was in the near term road map, no reason not to.
> Make profiles based on activities and swipe right or left on the activities you want to do.
Back in the early 2010s there was a dating site that was based around this premise. You would post a specific activity and see if someone wanted to join you for it. I didn't personally use it (as intended) but it was a great place to get fun date night activities with my partner. I think they realized that use case (date night planning) and eventually added this as a feature. I don't remember the name of the site unfortunately, someone else might. I would be surprised if it's still around.
> Out of all the AI-related tools, generative art frontends are probably the thing most likely to radically change and improve in the next few years.
The difference between UIs is actually not very relevant today; by now the generic workflow for complex scenes is more or less obvious to anyone who spent time with SD.
- Draw basic composition guides. Use them with controlnets or any other generic guidance method to enforce the environment composition you want. Train your own controlnet if you need something specific. (lots of untapped potential here)
- Finetune the checkpoint on your reference pictures or use other style transfer methods to enforce the consistent style.
- Use manual brush masking, manually guided segmentation (ex. SAM), or prompted segmentation (ex ClipSEG) to select the parts to be replaced with other objects. The choice depends on your case and need to do it procedurally.
- Photobash and add detail to the elements of your scene using any composition methods you have (noisy latent composition, inpainting etc) with the masks you created in the previous step. Use advanced guidance (controlnets, t2i adapters etc)
- Don't bother with any prompts beyond very basic descriptions, as "prompt engineering" is slow and unreliable. Don't overwhelm the model by trying to fit lots of detail in one pass; use separate passes for separate objects or regions.
- Alternative 3D version: build a primitive 3D scene from basic props (shapes, rigs). Render the backdrop and separate objects into separate layers as guides. Use them with controlnets & co to render the scene in a guided manner, combining the objects by latent composition, inpainting, or any other means. This can be used for procedural scenes and animation (although current models lack temporal stability).
As long as your tool has all that in one place, it's a breeze, regardless of the UI paradigm (admittedly auto1111's overloaded gradio looks straight out of a trash compactor nowadays). I expect 2D/3D software integrations being the most successful in the future, as they already offer proven UIs and most desirable side features. The problem is that in the current state SD can't do much in the production setting, it's not a finished product - so there's not a lot of interest in software integrations just yet.
I don't see it quite like that. It may be true that many employers are looking for postgrads. I don't know. But since practically _every single company_ large or small wants to know how or if they can take advantage of AI right now, there are plenty of opportunities to go around for anyone who has any knowledge or skills in generative AI.
And by far the most deployed technologies, even for postdocs, are going to be basically off-the-shelf or somewhat fine-tuned existing models via API or tooling being used for text and/or image generation for specific applications. In other words, things that in no way require the degree. That isn't to say that having real machine learning specialists isn't desirable, but there is plenty of room for people who just know how to apply the tools rather than invent new ones.
Actually there are plenty of "academic" papers coming out from people with advanced degrees that are mainly just applying GPT-4 to write code in some DSL to solve some category of problem. That stuff doesn't even require basic understanding of neural network architectures.
The latest models are so powerful it seems to be having a democratizing effect rather than increasing the class divide. From my perspective.
I've found myself resisting posting comments on Reddit recently - things like answers to simple questions people have about SQLite - partly because I don't want to cross the picket line but also because I realize that I was relying on a very light social contract that was in place there: I would share my knowledge for free, in exchange for which I knew that I was contributing to a larger dataset that myself and others could get back out again.
If Reddit are cutting off free API access, that social contract no longer holds. Why should I work to benefit their service if they're hoarding the resulting data and not making it available to me or people like me in the future?
I thought I was vanishingly rare in caring about this kind of thing, but given the mass Reddit blackout over the changes to the API apparently I'm not!
I stayed at my second job for nine years from 1999-2008. By 2008, I became an expert beginner. I asked myself for years why I stayed there so long. The current me, would have left April 2003 when I closed on the house I was having built.
If I do a retrospective, it is kind of clear:
I worked with most of the same people all nine years. My coworkers became my friends and they were the only constant in my life.
My managers were shady as hell though.
I was going through a horrible marriage (2002-2006) a horrible financial mess because by 2006 I had $500K worth of mortgages making $70K a year (because you could back then).
My skillset was outdated. I was maintaining VB6 code in 2008 - 7 years after it had been discontinued (and some old school C++/MFC/COM).
As bonuses got cut and raises were meager, I only made $8K more in 2008 than I did in 2001.
My head wouldn’t have been in the game enough for another job.
I learned my lesson with my next six jobs. I do resume driven development. I only worked at companies that were using in demand technologies. I jumped ship after two meager raises or when the bullshit/pay ratio got too high.
I kept my network strong.
But on the other hand, I’m going to “stay put” at my current job at $BigTech until at least my initial 4 year initial vest (and an interim refresher) is done and see what else is out there.
But figuring out how to incentivize people to self-determine their way into doing 100% of the things an organization needs to do is a real trick! Some work that still needs to be done just isn't great.
The typical approach is to use a combination of money and "skin in the game" (ownership) to cover those cases. And I think that's essentially the right approach, but this simply won't work for everyone.
And the thing is that it's super tough to separate out less fulfilling work into a distinct job role that can be more highly compensated and done by people who find that more motivating.
I think the best I've seen it work is for managers, who people already expect to be more financially compensated, to fill in these kinds of cracks. If there's something their team needs to do that just isn't getting done without them directing one of their reports to do it, can they do it themselves instead? (And maybe while they're doing it, they can be thinking about whether there is a way someone could build a tool to make it trivial to do the next time it comes up, and then they can probably get their team more excited about building that tool.)