post_below's recent activity
-
Comment on The real energy use of agentic AI in ~tech
-
Comment on Browser fingerprinting demonstration in ~comp
post_below Link ParentOn the other hand, consider things like the iphone. A particular iphone version provides huge swaths of users with identical hardware and software which makes fingerprinting significantly more...On the other hand, consider things like the iphone. A particular iphone version provides huge swaths of users with identical hardware and software which makes fingerprinting significantly more challenging.
The reason browser fingerprinting is useful at all is that the primary use case (outside of fraud/bot prevention) is advertising targeting, which doesn't need to be anywhere near perfect to be profitable.
-
Comment on Browser fingerprinting demonstration in ~comp
post_below Link ParentPossibly the reason you don't vibe with it is that the whole project, including most of the prose, is vibe generated.Possibly the reason you don't vibe with it is that the whole project, including most of the prose, is vibe generated.
-
Comment on The real energy use of agentic AI in ~tech
post_below LinkI love that the author took the time to do these estimates since there's so little non-vibey data on this topic. As always when it comes to climate change: Individual impact will not solve it...I love that the author took the time to do these estimates since there's so little non-vibey data on this topic.
As always when it comes to climate change: Individual impact will not solve it whether it's driving an electric car or avoiding LLMs.
I'm not saying that people shouldn't reduce their footprint, I reduce mine in all sorts of ways. I think it sends an important message to society at large about what our values should be.
But policy, regulation and infrastructure are the actual solutions, which we can see all over the world in countries that are successfully reducing emissions.
When it comes to AI labs and datacenters: Make them fully offset their impact. Make it expensive not to use renewable power, enough that it makes economic sense for them to fund new green energy development in some cases. Same for water use, it would only take a nudge to make recirculating cooling the right financial choice.
-
Comment on The magic is in the language in ~tech
post_below Link ParentIt's an interesting topic. Assuming I'm reading this right, I want to clarify a couple points. One of the key features of LLM agents is that they can be useful even outside of the direct context...It's an interesting topic.
LLMs would be useful, if they have been trained on the context they would have been used in. That means it works on the generic queries, but on the domain specifics LLMs either will not have enough training data, or will contort the "same words" from its training corpus, forcing a language from an aggregate of contexts
Assuming I'm reading this right, I want to clarify a couple points. One of the key features of LLM agents is that they can be useful even outside of the direct context they were trained in/on. They're more useful, for some things, when you're closer to the center of the training distribution but they can do something that looks like generalizing, some of the time, outside of the distribution
The reason for this appears to be that there's is more information encoded in the token relationships than you might imagine. One way to look at it is that words have both independent meaning and meaning that changes based on the context. Which means both that a word has self contained meaning that exists independently of any training data, and that dynamic meaning can be built on the fly out of word relationships as you write them. Because this is encoded in the weights, when you're running inference, meaning can survive into space that isn't specifically in the training. Even when there is no reference (so to speak) the model can blaze a path through the dark built out of token relationships and the meaning they encode. The words and the reasoning are the same thing, so they can go uncharted places.
It starts to fall apart at some point, but with a large enough model it can make it pretty far into the dark before that happens. That's especially true when it can generate feedback and incorporate the results into its inference, which is one of the defining characteristics of agents. The answer doesn't have to be in training for a model to find it as long as there's a way to test and iterate. In domains like code, math, biology and physics, deterministic testing is often easily accessible.
-
Comment on Share your favourite dal recipe! in ~food
post_below LinkLooking at the recipe that @phoenixrises posted, it looks great. You can do it a little quicker though, and still end up with great dal. Here are a few things I've learned... Butter/ghee and cream...Looking at the recipe that @phoenixrises posted, it looks great.
You can do it a little quicker though, and still end up with great dal. Here are a few things I've learned...
Butter/ghee and cream are key. Mostly that's just about fat and viscosity, so if you want it vegan instead you can use coconut milk (full fat with cream) and a neutral oil like refined coconut or avocado. Some recipes call for adding the coconut milk towards the end, similar to how you might do it with cream. I like to add it earlier, it helps mellow out the coconut taste.
You don't actually have to toast whole spices, or make your own garam masala. It does make a difference but it's pretty subtle. A good pre-made garam masala blend and a few other jarred spices gets you at least 90% there. Add the spices at the end of the veggie saute step to toast them a little.
Speaking of the veggie saute step, since the linked recipe doesn't include veggies... Consider sauteeing, at the least, finely chopped or grated onions in the recipe's called for oil before adding the other ingredients.
You can pre-cook the lentils in an instant pot (pre-soak totally optional) to save time.
The truly important ingredients are black lentils, some form of tomato (paste, diced, strained, etc..), cream/coconut milk, butter/oil, garlic, ginger and garam masala. You can just kind of experiment with everything else until you dial it in. I like to be conversative with the spices for dal, too much can distract from the staple soul food goodness of a fundamentally simple dish.
Oh and mash the lentils a bit towards the end with a potato masher or immersion blender, it improves the emulsification, texture and mouthfeel. You want to leave the majority of the lentils whole though.
-
Comment on What JavaScript framework should I use? in ~comp
post_below Link ParentThe way I see it, the main selling points of frameworks are standardization and compatibility. They make it easier to hire a [framework] developer to replace the previous one and they come with...The way I see it, the main selling points of frameworks are standardization and compatibility. They make it easier to hire a [framework] developer to replace the previous one and they come with varying levels of built in patterns that help stop interchangeable developers from making messes. Once upon a time they also helped solve cross browser compatibility, but these days, for most use cases, that isn't near as much of an issue as it once was. The last selling point is also outdated: adding functionality that's hard to implement in, for example, vanilla css/js. It's been a long time since I've encountered something that's significantly easier to do with a framework than it is with modern css/js. YMMV of course.
It's true that you'll end up writing something like a framework if you build a vanilla UI, but whether or not that takes more time depends on various factors. It definitely doesn't have to. And unless you really mess things up, your bespoke UI is going to be lighter weight than the framework alternative.
I'm not sure I see the maintainability angle. Are you thinking because many people aren't familiar enough with vanilla and so have a hard time understanding or writing it in a logical way? It certainly doesn't demand any bad patterns by nature.
Something I think it's easy to forget is that html/css/js essentially is a framework. One that's been iterated for the use case (websites/apps) for 30 years.
Not that frameworks aren't useful in lots of cases but, as nothing more than language constructs, they don't add as much as I think people sometimes assume.
Also: coding agents. The time to do X in a framework has now become basically identical to the time to do X without the framework. The remaining difference is load time on a slow connection and resource usage on a cheap device. In that context frameworks are just adding weight. Provided of course that the coding agent is in the hands of someone who is fluent in the underlying languages. Otherwise it could potentially be better to use a framework you're more familiar with, or with more good idioms than bad in the training data.
Anyway, I appreciate the digression, but my curiousity was more specific to @delphi's situation. It seems like an ideal opportunity to hand roll the UI instead of spending that time learning a new framework. Especially when a frontier model is almost guaranteed to give you a solid answer to every version of "how do I do this thing in vanilla js" you're likely to encounter.
-
Comment on What JavaScript framework should I use? in ~comp
post_below LinkWhy do you need a framework? HTML/CSS/JS can do all the things, but it sounds like you probably know that, so I'm just curious.Why do you need a framework? HTML/CSS/JS can do all the things, but it sounds like you probably know that, so I'm just curious.
-
Comment on 52% of Americans think their personal data will be breached. They're probably right. in ~comp
post_below LinkThat means at least 48% of Americans don't know their data has already been breached.That means at least 48% of Americans don't know their data has already been breached.
-
Comment on Four top Google AI researchers form new startup in ~tech
post_below Link ParentContext: Jeff Dean facts (which I just learned about) are a riff on Chuck Norris jokes. Highlights:Context: Jeff Dean facts (which I just learned about) are a riff on Chuck Norris jokes.
Highlights:
- Jeff Dean's PIN is the last 4 digits of pi.
- There is no Ctrl key on Jeff Dean's keyboard. Jeff Dean is always in control.
-
Comment on Inside Google’s $200bn Wall Street finance machine for Anthropic in ~finance
post_below Link ParentAlong with the crippled chatbots we'd also have a huge crash in global markets. A lot of industries are now looking healthy from a market perspective only because of AI related...Along with the crippled chatbots we'd also have a huge crash in global markets. A lot of industries are now looking healthy from a market perspective only because of AI related spending/investment/fomo
Whether that would cause a long lasting recession is anyone's guess, markets don't seem to follow the same rules they once did. At the moment these companies are propping up markets to as significant a degree as investment banking has in the past.
A crash, and a recession, might be the right thing in the long run, but there would be a lot of collateral damage to everyday people.
-
Comment on The magic is in the language in ~tech
post_below Link ParentThe relationship between cognition and language deserves its own thread because there's a lot of debate. I think some version weak linguistic determinism is self evidently what's happening, but...The relationship between cognition and language deserves its own thread because there's a lot of debate. I think some version weak linguistic determinism is self evidently what's happening, but the relationship between language and thought is so complex, and so hard to study, that it will be an open question for a long time.
-
Comment on As Reddit stock falls, CEO questions value of Google’s AI Overviews in ~tech
post_below Link ParentNot only is it walled off from the rest of the internet, it's controlled by a for profit company on its way to an IPO that will eventually enshittify.- Exemplary
Not only is it walled off from the rest of the internet, it's controlled by a for profit company on its way to an IPO that will eventually enshittify.
-
Comment on The magic is in the language in ~tech
post_below Link ParentIndeed. And it really is wild how quickly it starts to become normal. I'd have to stop and think about it for a while to count all of the things that happen on a given workday that would have been...this tech works in a way that I would’ve called laughably implausible a decade ago
Indeed. And it really is wild how quickly it starts to become normal. I'd have to stop and think about it for a while to count all of the things that happen on a given workday that would have been laughably implausible not very long ago.
-
Comment on Big food versus the people in ~health
post_below Link ParentExactly, the business model is addiction and food was a great way to expand into a market where no one thought of the products as addicting. And because no one thought of the products as...Exactly, the business model is addiction and food was a great way to expand into a market where no one thought of the products as addicting.
And because no one thought of the products as addicting, influencing policy and regulation was comparitively easy. For a long time the idea that highly processed food and refined carbs were a serious health issue was something that pretty much no one wanted to hear. Big food just needed to keep it that way. That didn't really start to change until the last couple of decades and they're still working hard to make sure policy doesn't catch up with the science.
-
Comment on The magic is in the language in ~tech
post_below Link ParentBefore fine tuning (which includes RL), next token prediction is a reasonable mental model, but fine tuning changes things quite a lot. You're right that there's a lot of capability embedded in...Before fine tuning (which includes RL), next token prediction is a reasonable mental model, but fine tuning changes things quite a lot. You're right that there's a lot of capability embedded in the weights.
Even with just pre-training, you might be surprised by the response to your example from a large model because the training creates complex and layered associations that already look like something more than next token prediction. A large model with no RL or other post training steps might not get the answer right, but it would "know" that it was a math question and its "guess" probably wouldn't be terrible because even if it had never seen those particular numbers in a sequence, it has relationships between a lot of numbers, and the shape of a lot of math problems, in training.
-
Comment on Big food versus the people in ~health
post_below LinkAdditional note: The list of 9 companies in the article should actually be 8. Kellogg's is owned by Ferrero and Mars (Mars has North American cereals, Ferrero has everything else).Additional note: The list of 9 companies in the article should actually be 8. Kellogg's is owned by Ferrero and Mars (Mars has North American cereals, Ferrero has everything else).
-
Comment on The magic is in the language in ~tech
post_below Link ParentYeah extreme interpretations both ways. It must be something we don't understand!Yeah extreme interpretations both ways. It must be something we don't understand!
-
Comment on Big food versus the people in ~health
post_below Link[...] Similar to the story of big tobacco, oil, healthcare (in the US) and pretty much every other mature industry with enough money to influence the system.Bad diets kill millions of people globally every year, and lead to tens of billions of dollars in health costs, often placing the heaviest burden on communities who are least able to afford it. Children, with limited decision-making power, are particularly vulnerable to this.
Legislators and government officials have worked to rein in this public health crisis by enacting laws that require food companies to be transparent about their ingredients and limit the advertising of unhealthy foods.
In public, the world’s biggest and richest food companies such as Coca Cola, PepsiCola, and Mondelez say they want to be part of the solution. But behind closed doors, they have taken governments to court to delay, dilute, and derail public health laws, which the companies say violate their rights.
[...]
Of the cases brought by private companies where the plaintiff was identifiable, more than 1 in 3 came from just nine parent groups, led by Coca-Cola, PepsiCo, and Mondelez
Their actions are not only prolonging the public health crisis but also cost countries billions of dollars in both legal and healthcare costs. In addition, they have a chilling effect on policymakers that wish to better their citizens’ health but do not have the resources to engage in drawn-out legal fights with food companies.
Similar to the story of big tobacco, oil, healthcare (in the US) and pretty much every other mature industry with enough money to influence the system.
-
Big food versus the people
19 votes
That's true, in the US things are (hopefully temporarily) completely fucked federally when it comes to not destroying the environment. It's down to states, counties and towns, some of which are doing encouraging things.