Stock prices goes up for shareholders when the C-suite declares they are integrating "AI". This is a well made and short video about the strategy, but the short of it is: "AI integration is not for the sake of employees, but investor's stock price".
https://www.youtube.com/watch?v=6Lxk9NMeWHg
Pretty nice idea, also introduced me to Ente's service which features shared event albums and guest append-only uploads - exactly what I needed a few months ago and even considered building myself.
Even if the factor is off by 10, it's already too much chemical exposure. Also EPA limits can often, due to industry pressure, be set way too high. So better be save and throw away that spatula, which actually this article also reiterates.
==== From the article:
“However, it is important to note that this does not impact our results,” Liu told National Post. “The levels of flame retardants that we found in black plastic household items are still of high concern, and our recommendations remain the same.”
So if you’re keen on eliminating these chemicals in any amount, chucking the black plastic kitchenware is a start, even if not as effective as the erroneous calculation suggests.
>Even if the factor is off by 10, it's already too much chemical exposure.
What's the basis for this conclusion, aside from taking the original study's author statements at face value? Someone else replied to the study[1] and characterized it as
>Based on a worst-case scenario, you may be getting nanograms (billionths of a gram) of bromine or lead from your spatula, which is lower than the amount that you get from eating fresh fruit
I did a quick skim of the atlantic article[2] and noticed that mentions of exposure thresholds were strangely absent, despite the pains they took to mention how those toxic substances were in the plastics, and how they caused harm.
That’s if you limit the comparison to just Bromine. You are ingesting recycled plastic from TVs and Tires, which also contains; Phthalates, Bisphenols, Polyvinyl Chloride, Styrene, PFAS, and several other chemicals shown to cause cancer or have an unlinked but possible cause of cancer (we don’t know yet but there are alternative utensils not made of black plastic that we do know that don’t have the big ?)
>which also contains; Phthalates, Bisphenols, Polyvinyl Chloride, Styrene, PFAS, and several other chemicals shown to cause cancer or have an unlinked but possible cause of cancer (we don’t know yet but there are alternative utensils not made of black plastic that we do know that don’t have the big ?)
You're probably ingesting all of that on a daily basis already, even without black utensils. Without a sense of scale, it's impossible to make a rational determination on what to do next. PFAS is in tap water as well. Should you stop drinking water?
You’re right that many of these chemicals are already present in our daily environment. But that doesn’t mean every source of exposure is equivalent or that we should stop caring about reducing it where we can. Think of it this way: if you know you’re going to be exposed to some harmful substances, it still makes sense to lower your overall intake whenever possible. It’s not about achieving total elimination, which might be impossible, but about minimizing unnecessary risks.
Eliminating known sources of contamination by (1) using safer utensils or (2) using water filters are straightforward steps that decrease the amount of these harmful chemicals you consume.
Even if we can’t quantify the exact benefit to the last decimal place, making such changes generally comes at low cost and might offer meaningful advantages over time. Why not err on the side of caution when the trade-offs are so small and the potential health benefits may be substantial?
That makes complete sense. For me, the time investment is totally neutral. I would buy a water filter anyway to mitigate other harms. Using stainless steel utensils vs plastic ones doesn't cost me any time at all. So I have risk reduced with 0 time investment.
Thank you for sharing that information about aluminum and brass cookware; I wasn’t aware of those details before.
The study you referenced does confirm the findings for aluminum and brass, but it also notes that no stainless steel cookware tested released enough lead to exceed childhood or adult Interim Reference Levels (IRLs). The testing involved placing vinegar—an acidic substance—and leaving it there for 24 hours, yet even under these conditions, stainless steel did not surpass IRL thresholds for lead.
I think what you are saying here is that avoidance of lead even below threshold is important and that plastic or silicone could have even lower amounts of lead in it. Which is fair and I think important to realize there is no perfect answer here.
In terms of cooking utensils silicone or bamboo might be a better choice than stainless steel. I don't think I've ever seen brass cooking stuff. I've often seen brass bowls with stickers warning that they aren't food safe.
My spouse had already mostly eliminated black plasticware from our kitchen long before this article came out anyway.
From a usability and quality perspective I would suspect that many on HN could afford the marginally more expensive, higher-end alternatives that will last longer anyway.
The idea that we should have ever turned recycled electronics, tires, and other non-food safe materials into food related implements has always been dubious on its face. It’s a scourge.
> Even if the factor is off by 10, it's already too much chemical exposure.
Source, please. Drinking one gallon of water a day is perfectly safe. Drinking 10 is not. 500 mg of acetaminophen is fine, 5 grams is not. 2 beers is reasonable, 20 is not. Factors of ten are pretty large safety margins.
The thing off by a factor of 10 here was not the level of the chemical, it was the calculation of the recommended safety level.
Also, different substances follow different safety curves. A 10x difference in acetaminophen takes you from very safe to very dangerous. A 10x difference in lead takes you from less dangerous to more dangerous; there isn't a "safe" dosage of lead.
> The thing off by a factor of 10 here was not the level of the chemical, it was the calculation of the recommended safety level.
Doesn't matter to the point I was making which is that a factor of ten is significant enough that you can't just assume what happens at X level means the same for 1/10th of it or 10x it.
I’m pretty sure take one lead molecule per day would not cause any detectable harm, even 1 additional nanogram likely too. So there a safe dosage of lead it is just hard to determine where the line between safe and unsafe is.
If you are able to detect lead intake in a person, and are unable to detect harm in that person attributable to that lead, that would be news to the scientific community and you should publish it.
A typical person seems to
breath between 8-9 cubic meters of air per day. So 1.2-1.3 micrograms of lead per day via oral inhalation seems perfectly fine per the EPA.
And the body does excrete lead, just slowly.
actual zero anything is nonsensical in the real world.
> > I’m pretty sure take one lead molecule per day would not cause any detectable harm [..] even 1 additional nanogram likely too
> "You're pretty sure" isn't unfortunately scientifically admissible evidence
From those two comments I feel obliged* to point out that one atom of lead - or indeed one atom of any element - is really a very, very, very small quantity...
A quick refresh on just how many atoms an element there are in one gram of that element might be in order.
> There is no known safe dosage of lead
I'd happily ingest a lead atom if you can prepare one for me.
Q: Is there a safe known dosage of ionizing radiation? I ask because I'm flying later today...
Yes, I'm familiar with exactly how small an atom is.
Identifying a single lead atom, and measuring harm caused by ingesting a single lead atom, are beyond the capability of our tools at the moment.
You can form a hypothesis that one atom would cause no harm, I can form one otherwise, but until we're able to quantify that then these are just hypotheses.
Among lead levels we are able to detect and measure the quantity of, there's no known safe level. And there very well may be points below which the harm caused, is so small that it's some people consider it an acceptable risk, as with your airplane and radiation example.
My point was that ‘no known safe level’ is not the same as ‘no safe level’. And it is not known because it is both hard to detect negative impact of small doses and because no one would spend money on such work when there are many more important problems.
It doesn't, obviously, since it's not possible to prove a negative. Much like there's nothing that proves that you won't begin to be able to fly if you stand on one leg long enough.
If what you've been shown so far is insufficient evidence for you, then I'm afraid you're on your own. I don't need to play the "how about this" moving goalpost game.
How do you know that exposure to comments by people named “cwillu” does not lead to cancer by some delayed chain reaction that takes decades to become visible?
Looks great, but this bug keeps from enjoying it: On firefox/macos/desktop, clicking on a POI on the map opens the sidebar but then redirects the whole page to the wikipedia entry.
"We utilize the last hidden state of the LLM as a representation of the reasoning state (termed "continuous thought")."
Could someone explain the last hidden state of the LLM ? What it shape is and how it is normally used - and why it hasn't been used yet to augment the next input? (which seems logical)
The last hidden state is just the output embedding after N residual layers, e.g. input embedding + res1 + res2 + ...
There's typically an "unembedding layer"/"classification head" that uses this hidden state to produce a softmax distribution over the LLM's vocabulary. In this case, we can think of this as "snapping" the hidden state into a single token and feeding that token into the next position of the autoregressive LLM.
In this sense, the last hidden state _does_ augment the next input. The authors simply propose directly feeding this hidden state into the next step rather than reducing it into a single token—thus, reasoning in continuous latent space rather than discrete token space.
Moreover “snapping” the hidden state to a token is akin to quantization. It’s lossy. By staying in latent space the model can “reason” at “full resolution” without discretization noise.
Sometimes discretization introduces interesting behavior though. Compare for example the logistic map and it's chaotic regime with the simplicity of the logistic ODE. Another example would be quantum mechanics compared to classical mechanics and determinism. The Poincare Conjecture was only interesting for n=3 due to too much connectivity in higher dimensions. Wouldn't it be interesting if consciousness only arose in such a discretized form, a case of incidental complexity and chaos introduced as the result of topological non-triviality from quantization?
Don't forget, non-linearity is fundamental to the whole process, otherwise you'd just have one large linear transformation. Maybe there's a similar role for discretization? :shrug:
Useful information about conceptual relationships and procedure can be captured in the LM head, so there is also potential lossiness when short-circuiting it.
Embeddings aka the last hidden state are the mathematical representation of an input of the model before a separate model (usually the decoder) translates that hidden state to a next token (the generative part in generative ai). Normally, the this step repeats over and over. This novel approach introduces re-using the last hidden state as if it was a token that has been generated thus "evolving" the hidden state over each iteration.
The way the recurrence in this method works -- ie, using last LLM hidden state at previous time step as input token for the next time step -- isn't directly compatible with how recurrence/autoregression is typically handled during LLM training. One of the major strengths of transformers is that they can be trained for recurrence/autoregression (which have sequential dependency) using convolutions (which are embarrasingly parallel). The proposed method requires introducing some sequential dependencies during training that could otherwise be avoided using "causal masking" and convolutions to enforce the correct dependencies between time steps in a sequence. Introducing these sequential dependencies makes training a lot slower.
tldr; the method requires training in a way that loses one of the major benefits of transformers, but maybe in some scenarios that loss is worth it.
I hope other countries take action soon. It's deeply irresponsible how we allow advertisements and Big Sugar/Fast Food companies to exploit colorful cartoon characters and misleading health claims to hook people—especially children—on excessive sugar and fat consumption. This not only fosters unhealthy eating habits but also conditions them to crave specific branded flavors from an early age.
> Parent commenter didn’t write anything to the contrary;
Did I imply that? Are "colorful cartoon characters" the part that's "deeply irresponsible" or just the misleading health claims? If you prevent the latter then you don't have anything to attach the former to.
> I really loathe these sort of “gotcha” comments.
I don't get what part of my comment was the 'gotcha'
Isn’t the issue there the claims of “fruit flavours” easily misinterpreted as having actual fruit and including vitamin c prominently when it’s just fortified.
You've moved the goalpost! No, the issue is that fruit flavors and cartoon characters are abused to appeal to children. That is what the original comment said and which you only half replied to! The video provided to you was an example of such - cartoons and 'fruit flavors' to hook kids on wanting sugary cereals for breakfast.
> No, the issue is that fruit flavors and cartoon characters are abused to appeal to children. That is what the original comment said
That's not what they said. They said:
> It's deeply irresponsible how we allow advertisements and Big Sugar/Fast Food companies to exploit colorful cartoon characters and misleading health claims to hook people
I don't think the cartoon characters is the part of the problem to address. And I think that aspect becomes irrelevant when addressing the actual problem. (And 'Fruit flavours' was my addition and question not theirs)
What part of my comment "only half replied"? I asked two questions in relation to the comment. I mentioned "colourful cartoon characters" and "fruit flavours" because those are examples of things that "appeal to kids" even though they have wider audiences. Canada (and other countries?) want to ban flavours from e-cigarettes because of the appeal to youth (and limiting it to just mint, menthol and tobacco). Why should an adult be limited to mint as a flavour for an age-controlled product? Why shouldn't an advertiser be allowed to use colorful cartoon characters?
Moved goalpost? Aren't we saying the same thing here? The video, and similar breakfast cereals, are perfect examples of the actual problem. It uses the combination of cartoons and false health claims. The false health claims are the parts that are problematic - that's the part to eliminate. The informational part of that video (and the box information design!) is meant to appeal to (read: mislead) adults, not kids.
Allowing false health claims for unhealthy products is an issue. Miseducating parents and adults about what's healthy is an issue. Suppressing how terrible excess sugar is for children and adults is an issue. Allowing lines like "how about fruit flavours"? (in video) and showing pictures of fruit is an issue. Allowing producers to hide how unhealthy something is by saying it's "healthy when part of a balanced breakfast" is an issue. Even now, nutrition information or prominent labels for cereals can include the values (eg protein) when one consumes it with milk!
Change that and what does that video or other ads for cereals become?
The purpose of the commercial is to create demand where there otherwise would not have been. Do you think that demand comes from the parents or the children? Which parts of the commercial do you think are most important for fomenting this demand?
I completely share your frustration with filler content and the tedious hunt for actual "news." (It speaks volumes about the state of the news industry—declining subscriptions, the normalization of clickbait, and a focus on quantity over quality.)
However, relying on AI as a solution has its own pitfalls: Even state-of-the-art models frequently generate inaccuracies and hallucinations, which raises questions about whether AI truly adds value if the extracted "information nugget" truly is what the original's essence is about or just another layer of BS.
Yeah. I put URLs I'm 100% sure weren't generated by AI spammers, and got out a third rate AI generated summary that skipped most of the detail. I put in URLs consisting entirely of obviously AI-generated shilling and got brief AI generated shilling. I put in the HN homepage and the text got longer ....
reply