CM Workshop · Beliefs Analysis

Next steps (updated 6 July 2026)

We're sharing this analysis now rather than waiting any longer. Here's what happens next — both for workshop participants and respondents, and for the additional forecasters and modelers we're now inviting.

If you took part in the workshop or already submitted beliefs

• This page is an internal draft summarizing the beliefs gathered so far. Please check whether your own views are represented fairly — corrections are welcome by reply, in the collaborative workshop document, or via the Hypothes.is annotation banner on this page.
• We didn't want to delay sharing any further. In parallel, we're inviting a slightly expanded group of forecasters and modelers — including people without ties to cultured meat, and some who have previously given more skeptical or negative forecasts — to share their beliefs too, for balance and dialogue. This helps balance the picture (admittedly somewhat ad hoc): people without direct CM links may have been less likely to attend the conference in the first place, so their views are under-represented here. We think this is important for getting at the cruxes and making real progress in public understanding.
• We now have two independent responses in this expanded round: Paul Wood (June 27) and one anonymous economics response (July 5). These were submitted after the workshop date, but they should be treated as pre-synthesis / pre-discussion-exposure independent views: the standalone form deliberately did not show the workshop estimates, model output, or meeting synthesis before asking for the respondent's own forecast. This is useful but still thin. Once we either have a few more independent responses or decide to close this round, we'll circulate an update to everyone: the revised synthesis plus the key apparent cruxes and discussion points.
• At that point — when there's a richer, more balanced picture to react to — we'll invite you to revisit and update your own beliefs. We expect that round to be more fruitful than asking for updates now.
• We're also considering additional live events (see below).

Technical sub-questions and expert distribution questions. Only respondents who answered each question are shown.

Process mode (E5) — % of production by mode in 2036

Only Elliot Swartz provided E5 values (fed-batch 20%, perfusion 50%, continuous 30%).

Discussion responses (open text)

Individual responses

All non-test submissions (includes one anonymized off-form response). Sorted by CM_01 estimate (low to high; no-estimate at end). Click to expand.

Methods, notes and annotations

Statistical approach

Respondent distributions: Cost is represented on the log scale because it is positive and often right-skewed. For the recommended linear mixture, a respondent with median m and 80% interval [lo, hi] gets a two-piece log distribution: the lower spread is ln(m/lo)/1.282 and the upper spread is ln(hi/m)/1.282. This preserves all three stated quantiles even when the interval is asymmetric. Respondents without an interval are assigned a user-selectable symmetric log-scale σ (0.3 = narrow to 1.8 = very wide).

Three aggregation methods are available:

• Linear mixture (recommended descriptive pool): F_agg(x) = (1/n)·ΣFᵢ(x). Each respondent contributes equal weight and asymmetric intervals are preserved. The result can be multimodal. Conceptually: "give equal credence to each respondent's full stated distribution." This is descriptive, not a claim that the group forms a calibrated posterior.
• Geometric pool (symmetric approximation): log f_agg = (1/n)·Σlog fᵢ + C. The closed form requires compressing each response to a symmetric log-normal. Location is the precision-weighted mean of log-medians and width is √(n/Σ(1/σᵢ²)). Narrow intervals therefore receive more influence, which is not automatically desirable: confidence can reflect overconfidence or shared evidence rather than accuracy.
• Product-of-densities illustration: Uses the same symmetric approximation but width 1/√Σ(1/σᵢ²). It becomes dramatically tighter. Calling this a Bayesian posterior would require each submitted distribution to be an independent likelihood with a compatible common prior; the workshop respondents share literature and discussion, so those assumptions are implausible. It is included only to show sensitivity to a strong pooling rule.

The methods need not agree on location. In particular, the equal-weight linear mixture's median is generally different from the precision-weighted location used by the other two methods. Report the method whenever quoting an aggregate.

Geometric mean of point estimates: exp(mean(ln(mᵢ))). Appropriate for log-symmetric distributions; less sensitive to the highest estimates than the arithmetic mean. Shown alongside the precision-weighted geometric mean (where respondents with tight CIs get more weight).

Exclusions: Three submissions have all slider-controlled probability fields at exactly 50%. The beliefs form uses a touchedSliders mechanism (added in beliefs.js) that prevents untouched sliders from being submitted — but these submissions predate that fix. They represent probable no-interaction responses and are excluded by default. One test submission is excluded entirely. "FN" is a respondent-entered pseudonym or initials whose meaning has not been confirmed. That submission includes substantive text and does not look like an untouched default, but it has no interval; the control above exposes its influence as a sensitivity check.

Respondent categories

Categories are assigned based on name + affiliation. "Researcher" includes both academic researchers and nonprofit analysts (GFI). "Industry" includes company practitioners. Several anonymous respondents cannot be categorised and are labelled "unknown."

The two anonymous May 8 all-50% submissions are the only ones that appear to be "default" (sliders not moved). The May 4 anonymous submission has a consumer-perspective reasoning ("current chicken price in my country") and a very wide CI — it is flagged as non-expert and can be excluded.

2027 near-term question

The original workshop form did include a supplementary CM_01 question for December 31, 2027 (fields: cm01_2027_median, cm01_2027_ci_lower, cm01_2027_ci_upper, cm01_2027_reasoning), but workshop respondents skipped it. The independent form has now produced two 2027 estimates: Paul Wood at $93/kg (80% CI $75-$119) and one anonymous economics response at $150/kg (80% CI $50-$250). Treat these as a small independent-forecaster sample, not as a workshop update. The July 5 submission is anonymized because the response did not identify the respondent in the survey fields.

Update analysis (pre/post)

Two workshop-cohort post-workshop CM_01 responses have arrived: PersonABC and Andrew Stout. Exact timestamps now distinguish before-workshop, during-workshop, and post-workshop submissions, instead of treating all May 8 entries as pre/during. The independent round adds Paul Wood and one anonymous economics response, but these are post-workshop only by calendar date and pre-synthesis / pre-discussion-exposure in the relevant information sense. For workshop-effect comparisons, the dashboard excludes the independent cohort and still has no paired pre/post responses from the same person. When paired responses exist, the dashboard will compute and display the shift in μ (log-scale location) and σ (uncertainty) for each respondent.

Hypothes.is annotations

Public annotations: Install the Hypothes.is browser extension and annotate freely. Public annotations on this URL are monitored by the workshop Slack bot (checks every 2 hours on the Linode server at 45.79.160.157 via hypothesis_to_slack.py).

Private group annotations:

1. Go to hypothes.is/groups/new and create a group (e.g. "CM Workshop Internal")
2. Copy the join link and share with collaborators (Zhuoran, Tommy, etc.)
3. In the Hypothes.is sidebar, switch from "Public" to your group name before annotating
4. To monitor this group's annotations automatically, update hypothesis_to_slack.py on Linode with the group's API tag

Data completeness