These are key issues with ~traditional econometric (IO and quant. marketing) methods.

Field experiments (supermarket-level or at school cafeterias etc.) have less of an endogeneity issue, but some of these issues are still present (e.g., short term vs long term), and these are hard to implement at scale and cleanly, and have issues of their own (see the notes/discussion, and sketch these).

Hypothetical and small-value choice experiments and hypothetical discrete choice surveys have other important limitations (mention these, from the sources and discussion).

That suggests

moderate this. This is vague. and there are a few kinds of field experiments in addition to this, including price shift experiments (esp. Bray et al), although few if any involving PBA

not "though"

this is a bit generic, maybe not necessary

likely --> "may" ... tooltip some other possible explanations for this. (discussed in the notes).

This is probly too strong ... needs caveating and referencing and tooltips.

just "seem to have serious limitations, and existing estimates are 'all over the map'" -- hyperlink or tooltip and link https://forum.effectivealtruism.org/posts/3Eh8MbqLwFBsD7GK2/how-much-do-plant-based-products-substitute-for-animal#Existing_Research

--> Observational demand estimation (using scanner, retail, and macro data)

rephrase. More like an "revisiting an underlying question"

make this 'between different animal products' and 'e.g., chicken vs. beef vs. eggs...' -- relevant for AW when considering issues like the AW impact of meat taxes -- which might shift consumption from beef to chicken, with a higher AW burden -- mention this briefly with further details in a tooltip

mention the insights here? I'm not sure we'll put out htis video either; it's not something interesting to watch , I guess. It was mostly preparation and broad discussion.

this is a strong claim ("most under-modeled") -- what's it based on? Reasoning transparency please. Provide support and links to this, tooltips etc. I want to make sure this is well-backed before I post and ~"co-sign" it !

all claims need more direct supporting evidence ... quotes, links, etc.; tooltips are your friend

wait -- are you sharing the beliefs here? we didn't wnt to do that yet!

skip/remove this -- no one showed up

TMI -- put that in a tooltip, perhaps

you keep pushing this -- it's ok to mention but the way it's written it seems like you are calling them out for hypocrisy; too harsh, and there can be other explanations

Eliot: Feed conversion ratio, taste studies on inclusion rates, actual cost of equipment, capital needs; scale out or up?

"We know more about media costs already"

European space agency tender for CM in 2021!!

20-25K per week for normal 200l bioprocesss -- JM ... maybe 15k with discounts. [Jordi's quick take]

An industry practitioner says: Focus on what is practical, where can I produce it, what will it cost

Oana suggested that embryonic stem cells would have a consumer acceptance problem. I don't quite see why that would be (animal welfare wise, isn't it just a few cells here as the starter, not many calves)? #question ... would it otherwise be high-value

"most uncertain" is stated a little bit too strongly here. .. Perhaps "Arguably the most uncertain "

Double check: is this the only solution, or are there other approaches?

This is likely very oversimplified and presents concepts that are already well known to most workshop participants, but some may be less familiar with the full process.

some of the tooltips are not coming up -- like here!!

Note, it's doing the sampling in straight javascript

link to the learn/explainer page on the different types of potential GF innovation !

Another folding block that should be folded by default.

Also, this should probably be a folding block folded by default.

This probably should be a folding block folded by default.

This callout block is not rendering.

not bold. Too much use of bold in general

Make this a tool tip

Use less bold text. Italics are generally preferred

Up time as a share I assume

"don't help" seems slightly too strong -- see previous notes #implement

Add as tooltip (condense a bit "What we think hydrolysates can do:

Replace the amino acid and peptide nutritional content that serum also provides (serum is ~60% albumin, plus amino acids, growth factors, hormones, lipids)

Possibly contain trace bioactive peptides that have modest stimulatory effects

Reduce the amount of growth factors needed by improving overall cell health and nutrition, so cells respond better to lower growth factor doses)

implement

replace with "basically" #implement

Give a tooltip here if possible "Growth Factor" - or just sell the word out the first time #implement

'reset adoption defaults' button is invisible -- too dark so too little contrast with the text.

Make reset defaults buttons more prominent throughout. #implement

Important -- bring in/merge updated content from https://valentinklotzbuecher.github.io/llm-uj-research-eval/

Meanwhile, link to that one for "the latest research"

doublecheck that these are NOT within the model context window

These numbers are being limited by Opus - we should do more evaluations in Opus and/or report the numbers just for the larger group.

extra ")" -- remove

has this been adjusted for the fact that we're comparing the machine LLM ratings to the average of human ratings rather than to individual human ratings? If not, it's would be an unfair comparison relative to human-human (averages have less dispersion), and there's a specific way to adjust for that.

Which comparison is this for? Is this human-human?

This seems like a great idea to me. Talked about it in the past, as well as giving evaluators some limited choice over which ones they would like to evaluate.

I take this to heart, and I'm particularly concerned that we just might not have the right team to do this well. That said, I'm not convinced yet that the space is fully covered in terms of rapid, expert, credible evaluations of this research. The alignment journal, for example, is thinking of acting a bit more like a traditional journal, and it also is limited in domain, as far as I understand, to alignment research and not other aspects of technical AI safety research.

But peer review of model cards and technical reports would indeed bring us into a more "technical AI safety" sphere, wouldn't it? And people were generally against that, as you see in the next section.

This synthesis seems to be overly abbreviated. It's not clear what is meant by this. What sort of research are we talking about? What methods does it use? Who produces it? What ways could it be evaluated? How could it inform impactful decisions and Pivotal questions for particular funders and policymakers?

I need to see some examples of this work to see if it's something that fits easily with our general approach and skill set

This seems quantitative, which is more comfortable to us.

I need to understand this work better. What is meant by this? What are some examples of such work? I am not sure if we want to extend ourselves to less quantitative/less empirical analysis (International Relations?) I'm also not against it, but it would be a bit of a move from our current approach and perhaps our comfort zone.

How is NBER overlooked? I don't understand. Maybe a quote from Tagat here would be helpful

I don't understand this. Why is it a "warning"? That would seem to be a good thing. We're not the ones doing the research; we're the ones prioritizing it, curating it, commissioning its evaluation, bringing feedback together, pushing it towards impact, etc.

Very much in our wheelhouse, and there is strong work. But does this really get at the GCR-relevant issues / the issues with the highest global impact?

This will be much easier with the use of coding tools. Standardisation will be less important/detailed

Concerns about AI alignment etc?

Hypothesis comments got lost, maybe in a page change because it changed the URL?

David van der Linden - Model validation Jacob Kozlowski - Financial modeling too strong ... they are involved

Clean Up the LaTeX here; it's not rendering right. #implement

Rename as "Model formulas" #implement

the TEA comparisons give different costs per liter, but I suppose they also give different cell densities. Should we interpret these as independent, fairly uncorrelated variables, or is the cell density more or less scaling with the price per liter in a way that using the actual cost per liter makes less sense?

crucial_uncertainty

should be May

the TEA comparisons give different costs per liter, but I suppose they also give different cell densities. Should we interpret these as independent, fairly uncorrelated variables, or is the cell density more or less scaling with the price per liter in a way that using the actual cost per liter makes less sense?

crucial_uncertainty

Add a link to this paper and to other papers within the table (a hyperlink) to save people time.

Claude: Suggest moving this entire discussion to the workshop site. This is a workshop-specific topic about how to frame CM_01, not dashboard educational content.

Replace with a one-line link: 'Regulatory implications for growth factor costs and how they affect CM_01 framing are discussed at our workshop site.'

Same applies to the 'Which market?' section further down. Together these save ~40 visible lines and reduce cross-site duplication.

Claude: Content trimming plan for this page (1,147 lines → ~700 target)

Proposed changes in priority order: 1. Fold FBS/serum-free section — historical context, no longer reflects current practice. One visible sentence + collapsed detail. 2. Move jurisdiction & 'Which market?' discussions to workshop — these are workshop topics, not dashboard content. Replace with one-line links. 3. Fold GF signaling diagram — educational but not cost-relevant. Keep the price/solutions tables visible. 4. Bold audit — reduce bold to headings + key numbers only (addresses multiple reviewer comments). 5. Trim Further Resources — link to workshop resources page for full list. 6. CSS fixes — larger diagrams, less whitespace around SVGs.

Optional: fold 2-3 SVG diagrams (cell banking, seed train) to reduce scroll length — the text already explains these steps.

Full plan at .private/content_trimming_plan.md. Feedback welcome here or in chat.

this should be specified somewhat by field

Also: Let users choose their own weights with sliders as well

is it a 'strong data set' etc

How do we prefer stated preference here? I think revealed preference and real choices are more credible

how literally is it using these 'standards' ? This bears some more expansion

caveat this if it's a conference paper

Make this more prominent.

add a category -- 'methodological and theoretical/technical issues to be evaluated ... rate 'evaluability'

provide tooltip quotes to check for hallucination

Not bold, avoid bold.

not bold. Too much bold in general

how is that 'optimistic?'

Note -- I want direct quotes from original sources (in tooltips) for important quant claims

K: scaffolding may not be a cost-effective option, low cell densities

Where is this used? I mostly see the earlier sources mentioned above.

The diagram is still too small. Make it bigger, please

not bold

can you have columns with the cost/kg of output for each of these, at each of the prices

keep this prominent (out of fold)

too much white space before and after image

give a tldr and the rest should be a folding box

Is that still true?! Source please

are they already doing this? If so, maybe adjust the wordking/emphasis here

give absolute cost/Kg too

whose?

who is this quote form? "achieving" needs a reference

at what cost/L?

"enormously for cost' Seems potentially too strong here. ... as we have suggested above, the estimated cost share for cell banking is less than 1% of the total cost.

Am I correct? If so, please moderate this.

For each step, give a 'tldr' and an estimated cost share, and have the rest be something folded by default, which they can unfold

references Humbird and Risner -- adapt htis to also reference work from GFI

missing bold label in diagram for this

Flag a note (more discussion in tooltip) about how this is the cost of pure cell mass, and early/ultimate products might be hybrid CM, plant-based, fungal etc. ... so this overstates the cost, in a sense

"Achieving $10/kg" is probably too strong. Maybe 'claiming the ability'? And do you have a link to this?

diagram below a bit small. Last item looks like a drumstick -- maybe make it look like a 'chicken hamburger' instead? (Because early products unlikely to have bones)

make these folding boxes with extra content in tooltips

The high cell density is in blue, but you also put "micros" in blue, which suggests the two have a link. I don't think that's what it is. I think the high cell density will reduce the media cost, which is in green, and maybe other goals like bioreactor and operating expenses so I'm a bit confused.

Diagram below does not really make sense. Is it a breakdown of the cost components or something having to do with levers that could make the costs go up or down substantially? This needs more clarity.

Hyperlink is not working so well

Top of this, or maybe on another page, it would be nice to have some sort of mosaic graph with different cost break-downs for different scenarios. Dividing up the cost into different components to get to total cost per kilogram, and then perhaps each of those mosaic elements could link to a different section explaining it.

This part of the model and also define what you mean by "binary switch" specifically in a tooltip.

I don't think these were mentioned anywhere either - Dash. Are these growth factors? If they're not growth factors, why are you discussing them in this section?

What is albumin? You did not mention it anywhere else in this document. Is it a growth factor?

Which of the growth factors do these "solutions" pertain to?

Where did you find the market data ?

It doesn't really matter what the price is per gram of inpuy . The question is, what is the likely price per kilogram of chicken meat output. add a column for this.

Tooltip example here, please.

Give listings, examples, and links/citations with tooltip quotes.

Why are you telling me about all these different kinds of growth factors? Do they all need to be used? Are they alternatives to each other? Have you defined what the terms in the "function" column mean?

And how much of them will need to be used per kilogram of chicken meat produced (or whatever weight we are standardizing things to here), what cost implications? Right to always bring things to this standard unit of cost per kilogram of chicken meat.

How does dollars per liter map into dollars per whatever unit of chicken meat we're using here? It's going to depend on the cellular density. I presume the cellular density is the same for these two types of media, or does one lead to much less dense cells?

To what extent is it clear that these can just simply be used, and to what extent is this still an important uncertainty? If it's clear that they can be used, We should make that clear-- to flag this so people don't think of it as still an important uncertainty. But we should look for more references here to be sure.

Try not to state things in a very prescriptive way. We're meant to be providing background information, not ordering people around.

Again, this really just seems too strong a statement to make. We need a little bit more epistemic modesty and reasoning transparency.

Another hovering "?" here

Where exactly does it come from though?

Maybe rewrite the headline to actually say that the total media cost is predicted to be out 40 to 70% of production cost, or whatever the numbers tell us. I don't need to expand it to get the headline result.

Any way you could cite more recent references?

Which ones give an explanation, maybe in a tooltip?

But why focus at research scale? Could you give what is projected to be a production scale?

Where does the supply come from? Are animals being killed here to produce it?

This might be a folding box. I'm not sure if it enters into the previous narrative. ?

So which of the above is this used for? It seems like it covers several of the above things you're calling "media". That's a little bit confusing to have this overlap of some sort.

Give me some excerpts from that page and explain what it means. You just linked to a sort of commercial page. It's not very helpful or easy to navigate.

A question mark comes up when I hover over this, but I don't see any tooltip explaining what it is.

Also, why are you talking about bovine serum if we're thinking about chicken here? At least you should explain the analogy.

Quantify this.

Maybe group the cheapest things together in one row unless there's some sort of environmental or ethical issue with them.

Tool tips should explain what each of these things are.

Not all costs, just this component of cost. Again, I want to know what share that makes up of the total to put this in perspective. It's only a minor share of total cost. It's not really a pivotal cost driver, is it?

Try to put these in terms of cost per unit of meat produced in a mature production process, and try to use the same units everywhere so we know how to compare each element and sub-element.

List these by order of estimated share of cost in a production-scale process. And give a rough estimate of those shares, and those should be on the same scale - expressed per unit of output, in the same units. Give a disclaimer, of course, that this is just based on one particular estimate, and you can link to the actual model.

What share of media costs are these in different models and reports? I thought this was possibly the largest?

Link this part of the model. Backlinks might also be good from the model to this explanation (here and everywhere else. )

Batch versus perfusion? You haven't given enough narrative here. I don't know why you're telling me this. Are these different bioreactor types, and if so, how does it map into the categories you just gave above?

This is a little confusing to me because what do you mean designed for food production? What is the standard food production use of this if not for cultured meat?

These images below are also too small.

These sources are all from 2021 or earlier - can't you find some updated cost figures?

What's the typical cost of the reactor transfer in an established, larger-scale production process? Would this still be a substantial share of total costs?

But isn't the sort of media that achieves higher density more expensive? I'm actually uncertain about this.

Source for the quote "often pharma grade?" Okay, you're relying heavily on Humbird here. Find some other sources, and I've heard that now most companies are using food grade instead of pharma grade. Look into that and discuss in tooltip footnotes.

For each phase, I want you to give some indication of the share of costs, in terms of the total cost per unit of meat, that this could potentially encompass, both at a small scale and at a larger scale.

The text is a bit crowded here, so the numbers overlap the words. Try to adjust to give it a little more space.

Give more continuous references, perhaps as tooltips, to where you are getting this information from about the process. Perhaps give citations with links and short quotes.

Link to this regulatory approval thing - how difficult/Costly is it to get that approval, or do we already have this for the important immortalized cell lines?

Doesn't really explain how the costs work. Ultimately, the banked cells are used up, correct? Are you saying that cell banking is just a tiny share of the cost here, if you end up using the whole batch, is that right?

Image is still a bit too small.

Are we trying to isolate and why? Maybe this is a tool tip? And what's enzymatic digestion?

You did not use the term "cell banking" in the flow chart above. This can be confusing when you change terms. We don't know what Maps to what

Give the name of the paper and a tooltip, and also explain what aspects of these claims the source provides, perhaps with quick quotes.

Explain, perhaps in a tooltip, why the similarity is helpful here. That I don't really know what these things mean (e.g., what does ng mean?)

Provide a tooltip/link to discussion from animal welfare advocates about this, perhaps on the EA forum.

Tooltip hover should show the name of the paper, et cetera.

Can you make an image without a bone in it that still looks like a piece of chicken meat? I don't think bones are happening any time soon in cultured meat.

What's a seed train?

Seems too strong. Media costs exceed GF costs in many formulations I've seen.

this seems a bit too strong from my reading. Media costs exceed GF costs in many formulations

I don't think this is for pure cell mass, this is for a hybrid product

folding box

a continuous variable I guess, not binary. Make this clearer to avoid confusion

Based on which parameters? the user-entered ones above?

Important -- are these means or medians of a distribution used in simulation or are these simple 'degenerate' numbers. Explain and signpost better

And do they affect the figures and graphs below? #important

Do these adjust as I change the sliders above? I changed the 'plant capacity' and I did not see any change here. What's going on? #important

only these 3? Does the model incorporate other info?

which model? Does this adapt dynamically?

give some tooltip explainers here

make the graph below a bit bigger

Wait -- adjust this to consider/note the CM inclusion rate (%) and cost of plant-based or mycoprotein ingredients ($/kg)

Where did the slider range and starting value come from? Explain in a tooltip

Have a separate number for each of these boxes (slightly less prominent) for 'hybrid product cost/kg' ...

User should be able to input a 'CM inclusion rate (%) and cost of plant-based or mycoprotein ingredients ($/kg) as parameters', and this should do a simple auto adjustment.

First just a simple adjustment, and later we make this part of the simulation model.

Also allow user to switch this 'hybrid product' on/off (box below 'Include downstream processing'

Caveat/note here about cost of producing the pure cultivated chicken cells, vs cost of the product that will have some percentage of these cells mixed with other (plant, fungal, etc.) ingredients.

why wouldn't you include these?

give some specific examples of cases where this is used

Note that for ~switching parameters, the model samples both from the probabilities of a switch to a different regime (a different discrete state of the world, e.g., a new discovery) and then, in each simulation, uses this probability to select a particular state. (word this better, and be sure I'm correct here)

no "!"

But atm we donm't have confidence in this model. I's more about fixing ideas and giving people a sense of what sort of modeling we want (and surfacing doubts and disagreements on this) so that we can productively collaborate.

Ideally, we'd also have a page/interface where people could 'build their own models' and we compare them.

This should probably be mirrored or linked a bit more on the man quote model page so people understand where it comes from.

Skip this last bit. Try to condense the content at the top a bit more for this page

Don't need this at the top of the technical reference page.

Explain why we targeted this 'cross category' substitution. We understand it's initially counter-intuitive. TLDR: Impossible Beef is a defined category with fairly clear pricing and high quality, but chicken consumption is more animal-welfare-relevant than beef. But we have variants of this that are more within-category. Put the TLDR in a tooltip and a longer explanation in a folding box.

Some more detailed explanation at https://forum.effectivealtruism.org/s/kazWBBYXm2Rvya3y2/p/3Eh8MbqLwFBsD7GK2#Why_focus_on_chicken_consumption_ and https://forum.effectivealtruism.org/s/kazWBBYXm2Rvya3y2/p/3Eh8MbqLwFBsD7GK2#Why_focus_on_Impossible_Beyond_Beef_

I'd link the Metaculus question right next to each question here.

Still add your annotations and let me know what you think about hypothesis as a format for collaborative discussion.

This feels a bit vague to me. Are there specific policies that would be affected?

remember -- we are not focused on cross-population comparisons for this workshop. It's more about 'which interventions yield greater welfare', which would generally involve differences in difference, ideally across comparable populations (but not always)

Where did the discount and time factor come from? Where did these definitional equations come from? I didn't think most emply estimated WELLBY measures considered multi-year collection or impact. And are they really discounting?

Cut this. I don't think it necessarily holds -- a lot of interventions impact mortality.

Add to footnote -- the 'incremental' WELLBYs may be captured by observing differences between comparable treated and untreated populations.

it's not really 'accounting' -- these are conceptual

A link to this would be helpful. The "Green Book". (I wonder -- how impactful has this actually been on British policy?)

Adjust this -- "If you prefer your response to remain anonymous, please use a pseudonym and try to use the same one consistently if you're providing multiple responses." If you are fine with internal recognition but don't want any public attribution, please let us know and share any other concerns in the field at the bottom.

I'm considering adjusting this one to

Consider the 'value obtained when using the best feasible measure for cross intervention comparison in contexts like the focal context'. What share of this value is obtained, in expectation, from using the simple linear WELLBY measure for all interventions? Please give your central belief, and 90% credible intervals"

-- with a slider that goes from zero to one, and two other sliders that allow that allow you to specify the lower and upper bound of the 90% CIs.

NotebookLM:

"they applied their methodology to nine prominent results from the happiness literature—including the Easterlin Paradox, the U-shape of happiness in age, the ranking of countries by happiness, and the effects of marriage and children—and showed that the standard conclusions in all nine areas could be reversed using monotonic (specifically lognormal) scale transformations. They argued that these reversing transformations were "plausible," claiming they were no more skewed than the U.S. wealth distribution

However, later work questions the plausibility of this. .

See https://www.econstor.eu/bitstream/10419/232657/1/dp13905.pdf

It will take more than 60 minutes, but a skim might be doable in 60 minutes.

Add "(rsvps)" -- Also update this, I think we have over 20 by now.

is this ggplotly? Shouldn't it be dynamic? I don't seem to be able to adjust it

These are probably not updated costs -- Update and implement

Not 'video', possibly some written content, or we can extract issues from his evaluation to ask Benjamin et al.