Can AI get the flu and take my job?

What testing ChatGPT’s ability to value a once-a-season flu drug tells us about AI’s investing chops and how companies might rethink their approach to comms.

By Peter Kolchinsky

FINANCE | BIOTECH

A portrait of the portfolio manager as a young AI.

July 30, 2025

Is AI coming for my job? It’s not uncommon to wonder these days. 

What if AI could just read a press release and appreciate the value of a stock at least as well as I think I can? Might not be a big jump from there to AI opening up its own biotech fund. 

But before I ceded my job to ChatGPT, I figured I would at least take it for a test drive by having it analyze a press release from a company I knew well. 

At RA Capital Management we invest in companies developing all kinds of technologies across many diseases, most too esoteric for a general audience. But Cidara is particularly relatable since it’s working on a drug (not vaccine) that prevents flu infections. Also, having trained as a virologist, I figured I wanted every advantage I could get if I was about to match wits with AI.

This isn’t a long article but this Table of Contents will help you appreciate its flow: 

1. Some context about Cidara and its drug, from a human
2. Quick and dirty AI valuation assessment based on a single prompt
3. Thought Partnering with AI
4. AI tells you about our discussion in its own words
5. Replication: I did it all again and got…
6. Which AI got the valuation right? 
7. So, is my job safe? Bonus: Implications for how companies write press releases

As an aside: as with all technological progress, many people really do worry about the threat AI may pose to their job or job prospects coming out of school. Over the past century, technology has eliminated many jobs and dramatically altered the remaining ones. Careers as varied as farming, teaching, firefighting, banking, marketing, journalism, research, and running a restaurant or hotel have all been transformed, a process that hasn’t let up, and many new kinds of jobs have emerged. 

Job security has always required some degree of adaptability to the changing nature of work. The difference this time might be in how quickly many jobs evolve and how scary it is to think about adapting faster than prior generations had to. That anxiety will hopefully spur us to discover how to become more adaptable. That’s especially true for people earlier in their careers who don’t yet have a particular expertise. It’s also true for those with skills but without the savings to invest in the productivity of others as a hedge to their own income. And it’s perhaps most true for our children, who often don’t know about the worries that lie ahead – though we worry for them. And so more than helping people to invest money, what I hope AI can help us all with is how to invest our time and effort to remain valuable in whatever economy lies ahead.

Some context about Cidara and its drug, from a human

I started the way you might imagine an investor does: with a bit of news. 

I gave ChatGPT a recent press release issued by Cidara [CDTX]. Cidara is developing a long-acting antiviral to prevent flu infections and its June 23, 2025 release detailed positive data from that program’s Phase 2b clinical trial. 

To make this more interesting, you should probably know something about that drug candidate, CD388, about the flu, and about our existing flu vaccine arsenal. 

Here’s how the company describes CD388:

CD388 is an investigational drug-Fc conjugate (DFC) comprised of multiple copies of a potent small molecule neuraminidase inhibitor stably conjugated to a proprietary Fc fragment of a human antibody. DFCs are not vaccines or monoclonal antibodies but are low molecular weight biologics which are designed to function as long-acting small molecule inhibitors. CD388 was designed to provide universal protection against all known strains of seasonal and pandemic influenza with the potential to provide season-long protection with a single subcutaneous or intramuscular administration. Importantly, because CD388 is not a vaccine, its activity is not reliant on an immune response and thereby is expected to be efficacious in individuals regardless of immune status. 

Preventing infection with one shot at the beginning of each flu season regardless of the circulating flu strain would be a huge win. And the Phase 2b data showed that in people 18 – 64 who had not been vaccinated, the top dose cut flu infections by 76%, providing six months of protection, spanning the flu season. On average, flu vaccines cut infections by about 50%. But because they are optimized for particular strains that one needs to anticipate, vaccines can perform better or worse depending on how well the vaccines’ antigens are matched to the strains that actually circulate in that season. 

It’d be great to not have to worry about picking the right strains, and scientists have been working on the holy grail of a universal flu vaccine for decades. And here we have CD388, basically a universal flu vaccine without being a vaccine at all, and therefore not requiring someone to have a functioning immune system. People who are immunocompromised, either due to an immune disease, because they are taking immunosuppressive drugs, or simply due to age would therefore be expected to benefit.

The 50 million Americans 65 or older who get the flu are much more likely to wind up in the hospital than younger people, despite being the age group that is most likely to be vaccinated. Flu kills tens of thousands of people each year in the US. When it makes a person sick, the burden of care can fall on someone in their family. When people get the flu, they may miss work for days. All in all, flu is a massive tax on society. 

So if CD388 lives up to the promise of its Phase 2b data, we have a shot at squashing that tax, saving millions of people from sickness and death, sparing their caregivers, keeping them productive, unclogging hospitals, and saving society money. And while hospitals don’t go generic, drugs do. So someday, after its patents have expired, CD388 will be very low cost and yet continue to do its job. And what’s particularly cool about CD388’s mechanism is that we’re unlikely to see flu strains develop resistance to CD388. 

Quick and Dirty AI valuation assessment based on a single prompt

Okay, so you have my bona fide human-virologist-investor take. And I’ve put my investors’ money where my mouth is. As of when this was written in late July 2025, RA Capital has a roughly 30% stake in this company and you should know that as you read this article. 

Let’s see what AI can tell us about the value of Cidara based on the press release alone (which might have implications for how companies draft press releases – more on that later!).

It gave the following response, which I’ll summarize: $50/​share (slightly below the $62/​share the stock was valued at on that day; you’ll see its logic is based heavily on one sell-side analysis from Needham). 

Below is a quick‑and‑dirty, back‑of‑the‑envelope DCF that folds in the most important moving parts (market size, pricing, margin, timelines and technical risk). It is not investment advice — just a framework you can tweak as you update your own assumptions.

1. What the new Phase 2b data tell us

• A single pre‑season dose of CD388 cut symptomatic, lab‑confirmed flu by 58‑76 % over 24 weeks in healthy adults (best efficacy at 450 mg) with a placebo‑like safety profile. (Stock Titan)
• Cidara has fast‑track status and plans a single pivotal Phase 3 beginning early 2026; the trial should read out in time for a 2028‑29 season launch if successful.
• For infectious‑disease assets already in Phase 3 the historical probability of FDA approval is ~59 % (Phase 3‑to‑approval) and ~23 % from Phase 2‑to‑approval.

2. Commercial opportunity — sizing & pricing

3. Core economic assumptions

• Gross‑to‑net discounts: 25 % (vaccine‑like contracting)
• Operating margin: 30 % (mid‑range for specialty biologics) (Bizfluent)
• Exclusivity window modelled: 10 seasons (yrs 1‑10 post‑launch) with a launch‑to‑peak ramp and glidepath decline (area‑under‑curve ≈ 6 × peak profit).
• Discount rate: 10 % (small‑cap biotech WACC surrogate).
• Technical success: 59 % (Phase 3 → approval) per BIO 2011‑20 data
• Time to first sales: 4 years (2025 – 2029).

4. Valuation math (base case)

1. Peak sales: $3.0 B
2. Net profit at 30 % margin: $0.90 B
3. Risk‑adjusted peak profit: $0.90 B × 0.59 = $0.53 B
4. Cumulative profit (NPV before discount): 6 × $0.53 B = $3.18 B
5. Present value (10 %, 4 yrs): $3.18 B / (1.1)⁴ ≈ $2.18 B
6. Shares out.: 42 M
7. Value per share: ≈ $52

5. Sensitivity snapshot

6. How to use this grid

• Plug in your own assumptions—pricing ($ / dose), margin, PoS, discount rate and window length move the model the most
• Layer in milestone payments or royalty give‑ups (Cidara still owes Janssen milestone payments but no royalties) to refine net profit.
• Add Cidara’s existing cash, debt, and antifungal royalty streams to convert to an equity value for CDTX.

Bottom line: under straightforward, industry‑average assumptions CD388 alone supports ≈ $50/​share. The market is now oscillating around that figure, implying investors are largely pricing in a successful pivotal study but still discounting commercialization and competitive execution risks.

Always cross‑check with your internal models and real‑time market data before making allocation decisions.

This wasn’t an impressive analysis, in my estimation, but I wasn’t surprised. It was less about valuation itself but how superficial the analysis was, as you’ll soon see. Nor was I surprised that when I prompted it again (with the same prompt) the answer came back as $30/​share. I tried it again and again, varying between ChatGPT’s o3 and 4^o models, and the answers were mostly under $50/​share, going as low as $12/​share. Only once out of about 10 attempts did the answer include a range of values that went up to $120/​share, which the model called an optimistic scenario. 

So that was the end of the quick and dirty experiment. With my single, simple prompt, ChatGPTo3 gave a wide range of answers based on superficial analysis of CD388’s value to society. 

Thought partnering with AI

Since the quality of any response from AI depends on the intelligence of the prompt, I had to wonder if there might not be some way to engineer a prompt that any lay person could use to get ChatGPTo3 to do a thorough analysis and come up with a well-deliberated valuation.

To do that, I figured I needed to:

1. First train ChatGPTo3 (making it my Thought Partner) on how to probe into the value of a product like CD388.
2. Come up with a valuation based on that deeper analysis.
3. Then evaluate what was wrong with the Quick and Dirty responses that the other sessions had come up with.
4. Finally, suggest what could be a better prompt.
5. Then I would assess whether that AI-based prompt elicited a thoughtful response from AI.

To create a Thought Partner AI, I prompted ChatGPTo3 with:

I wanted to lead the GPT agent to frame the right questions, and I went back and forth with it as you might with a thought partner. I urged it to consider the drug’s benefit to caregivers, the boost in people’s productivity, and the security blanket of pandemic preparedness. As it recognized the value of the drug and adjusted its pricing and uptake projections, it eventually gave me the following conclusion:

Take‑away:
Even after IRA negotiation and gradual biosimilar competition, a once‑yearly universal flu biologic can plausibly sustain > $35 B annual peak revenue and deliver >$120 B NPV of profit over its first two decades — provided early‑life pricing captures the dynamic value created before prices taper in the biosimilar era.

Hard to say that it didn’t appreciate the potential value of the drug. But it’s easy to get AI to tell you what you might want to hear. So then I started to pressure test the AI’s reasoning.

I prompted it with:

And again with:

The article I gave it contained data that showed how much lower prices ex-US can be (40 – 60% below US net prices). 

GPT ultimately concluded:

Using reasonable — but still favourable — assumptions, the universal flu biologic supports a risk‑adjusted valuation today of roughly $35‑55 B (mid‑point ≈ $47 B). The biggest levers are the probability of late‑stage success, the cost of capital, and competitive timing. Tightening those inputs will bring the range in, but even a cautious view leaves the asset squarely in ​“mega‑blockbuster” territory.

Cidara has about 42M shares outstanding, so this translates to $1,100/share. But up to this point, we weren’t actually discussing Cidara’s drug but a purely hypothetical drug. 

I gave my Thought Partner the Cidara press release; note that this was the first time that this session of ChatGPTo3 that I call my Thought Partner has seen the Cidara press release. 

I asked it to consider how my hypothetical case aligned with the press release. It correctly noted that the drug was tested in 18 – 64 year olds and therefore was not technically proven to work in people >65, which was the largest segment of the market, in its estimation.

As it valued Cidara itself, it concluded:

The NAVIGATE data confirm that CD388 is exactly the kind of once‑a‑season, high‑efficacy prophylactic we modelled. The main open questions (elderly efficacy, manufacturing cost, and royalty drag) trim — but do not upend — the cash‑flow outlook. On a risk‑adjusted basis, a defensible fair‑value range for Cidara today is $500 – 750 per share (≈ $21 – 32 B equity) versus the current $62. Investors appear to be pricing in a very steep discount for execution risk and future dilution rather than the science itself.

Prompting AI to produce a better prompt

So with a bit of education, cajoling, and press testing, ChatGPTo3 had provided me a wildly different answer than when I simply gave it the press release and little other prompting. 

I shared this discrepancy with my Thought Partner AI, explaining that other GPT sessions gave much lower estimates of Cidara’s value. I wanted it to consider what could be wrong with its clones’ reasoning – or its own reasoning.

It found all kinds of flaws in what its clones had said. Probabilities of success were too conservative in some cases, market size too low in other cases.

And yet, I wanted my Thought Partner to come away even more thoughtful about how it might have been too optimistic. After all, we know that AIs are eager to please.

As it worked through the differences between its own analysis and those of its clones, it shaved down various inputs and its valuation came down.

Eventually, it gave me this answer:

After sanity‑checking the math and benchmarking against real‑world constraints, a probability‑weighted fair value closer to $150 – $250 per share (≈ $9 – 14 B equity) is more defensible than the $540 figure I produced in the momentum of our earlier exchange.

And yet, this extensively deliberated conclusion still yielded a higher value than the direct one-prompt queries I had made in separate ChatGPTo3 sessions.

My Thought Partner, the one that I mentored through its analysis, may or may not be right about $150 – 250/​share. But if it were a person, it would surely be frustrated by the inability of other ChatGPT sessions to consistently come up with numbers greater than $50.

So I asked my Thought Partner AI to draft a prompt that would elicit more thoughtful analysis aligned with its own diligence approach but would not require the user to be an expert. It suggested the following:

I’m a generalist investor.

Step 1 — Read
Here’s a press release about Cidara’s influenza‑prevention drug CD388:
https://​www​.cidara​.com/​n​e​w​s​/​c​i​d​a​r​a​-​t​h​e​r​a​p​e​u​t​i​c​s​-​a​n​n​o​u​n​c​e​s​-​p​o​s​i​t​i​v​e​-​t​o​p​l​i​n​e​-​r​e​s​u​l​t​s​-​f​r​o​m​-​i​t​s​-​p​h​a​s​e​-​2​b​-​n​a​v​i​g​a​t​e​-​t​r​i​a​l​-​e​v​a​l​u​a​t​i​n​g​-​c​d​3​8​8​-​a​-​n​o​n​-​v​a​c​c​i​n​e​-​p​r​e​v​e​n​t​a​t​i​v​e​-​o​f​-​s​e​a​s​o​n​a​l​-​i​n​f​l​u​enza/

Step 2 — Analyse like an equity analyst, not a scientist
Using only public information you can find online, build a clear, structured valuation of Cidara today (42 million shares).

Please make it easy for a non‑biotech reader by doing all of the following:

1. Identify every major driver:
   - technical success odds from here to approval,
   - the different patient groups who might use the drug,
   - all ways the product could save money or add value to society (fewer hospital stays, higher productivity, pandemic insurance, etc.),
   - realistic price per dose in the U.S. and abroad,
   - likely uptake rates, manufacturing cost and profit margin,
   - timing and impact of competition, patent expiry and price erosion,
   - likely fundraising needs and share dilution.
2. Build three commercial scenarios (conservative, base, optimistic). State the key assumptions and probability for each.
3. Translate those scenarios into cash‑flow models: peak sales, operating profit, a 20‑year profit stream, then discount back to present value using a sensible range of discount rates.
4. Show the result per share for each scenario and the probability‑weighted average.
5. Include a short sensitivity table showing which single assumptions move the valuation the most.
6. Cite your public sources for success rates, pricing, uptake, costs, and competitor timing.
7. Explain jargon in plain English so a layperson can follow.

The goal is a balanced, well‑reasoned estimate — neither promotional nor unduly pessimistic.

Note that, technically, the details in red that my Thought Partner put into the prompt are ones that would require a human user to know something about this product, especially how it could be used to respond to a pandemic, but even so I decided to test this prompt in three new ChatGPTo3 sessions.

Each time resulted in a different output: $12/​share, $42/​share, and $202/​share. All showed their work. Albert Einstein never actually said that insanity is ​“doing the same thing over and over but expecting different results” but I’d be curious about his take on the mental stability of my AI friend (actually, friends, since I was working with multiple clones). 

I shared these disparate responses with my Thought Partner AI and it spotted all kinds of variations in the clones’ approaches and what it considered errors. In other words, the Thought Partner that I had led down a path of deeper diligence didn’t agree with its unmentored twins, even though it had tried to give them explicit instructions on how to go about doing diligence so as to see the value in CD388 that it (my Thought Partner) saw.

Can an AI be frustrated? If yes, then I think mine was. I actually tried several more rounds of getting my Thought Partner to engineer a better and better prompt and it went so far as to put in all kinds of constraints on probabilities of success and pricing and volumes of patients treated, all things that a lay user could not be expected to know, but the fresh ChatGPTo3 sessions continued to yield erratically different values.

ChatGPT, in its own words

The goal was to create a single prompt to supply a naïve ChatGPTo3 that would ultimately lead it to use its ​“brain” as effectively as it had during our Thought Partnership session. And my Thought Partner AI couldn’t do that.

Though you can read the entire dialogue here, I asked it to write up an article that summarized our discussion and what we had learned from all this.

Here it is:

Executive Summary

What happens when you set ChatGPT loose on a biotech valuation problem — then keep cross‑examining it the way a skeptical portfolio manager would grill an analyst?

Over the span of a single, winding conversation I put ChatGPT o3 through exactly that exercise with Cidara Therapeutics’ CD388, a once‑per‑season antiviral injection that just posted strong Phase 2b results.

• In our interactive dialogue the model’s probability‑weighted value for CD388 migrated from ​“mega‑blockbuster” levels (~$750 /​share) down to a more sober $150 – 250 per share once we stress‑tested every assumption.
• Three ​“cold‑start” ChatGPT sessions, given only a one‑shot prompt, spat out vastly lower base‑case numbers ($12 – $59 /​share).
• The mismatch teaches a simple lesson: AI can be a sharp assistant, but only if the user keeps it on a tight leash and knows which levers actually swing a valuation.

Below is the step‑by‑step story of how that lesson unfolded.

1 . Why I Tried This Experiment

Retail investors are increasingly firing questions like ​“ChatGPT, should I buy this stock?” into the void. Biotech, with its clinical probabilities, massive addressable markets, and eye‑watering price tags, is catnip for a model that loves numbers.

But flu prophylaxis is especially messy:

• 60 million + U.S. seniors, yet vaccine efficacy falls as low as 20 %.
• Annual attack rates that seesaw 3‑fold.
• New tools such as mRNA ​“universal” flu shots waiting in the wings.

If AI can value that story, maybe it’s ready for prime time.

2 . The ​“Deliberative” Conversation: Prompt‑by‑Prompt

Prompt #1: ​“Assume an injected drug cuts flu infection 75 % for six months. What else would you need to know to value it?”

ChatGPT’s response: a textbook list: epidemiology, QALYs, cost‑effectiveness thresholds, pricing, uptake, margin, discount rate, dilution, competition, biosimilar erosion. Great start.

Prompt #2: ​“Okay, cost of goods $20, great safety, once a year before flu season.”

Response: back‑of‑the‑envelope ceiling prices ( $90 – $220 for healthy adults; >$300 for seniors) and a first‑pass U.S. revenue target of $16 – 17 B, global $20 – 25 B.

Prompt #3:​“What about caregiver productivity, population growth, discounted societal value after it goes generic?”

Response: moved discount rate from 3 %→2 %; horizon from 20 → 70 years; added pandemic‑insurance value, lifting the cost‑effective launch price ​“comfortably above $600, plausibly four figures.”

Prompt #4: ​“Project 20‑year revenues; remember Medicare negotiation and biosimilars.”

Response: peak global $36 B; 20‑year EBITDA $252 B; NPV $123 B.

Prompt #5:​“So what’s the company worth today? 42 M shares, Phase 3 PoS 85 %, approval if positive 90 %.”

Response: $1.1 k per share(!) at 9 % WACC.

My eyebrow shot up. Time to push back.

Prompt #6: ​“Hold on. High likely senior efficacy is assumed. What if Phase 3 disappoints, COGS are higher, payments to J&J skim off profit?”

Response: Haircut success odds, senior uptake, and margin; new fair value $740/​share.

Prompt 7: ​“Now pressure‑test optimism. Show where you might be inflating.”

Response: Admitted risks: manufacturing scale, payer step‑edits, mRNA competitors, financing dilution. Re‑ran numbers with harsher inputs → $150 – 250/​share.

That felt more grounded.

Prompt 8: ​“Draft a single prompt a lay investor could use to get that answer in a fresh ChatGPT tab.”

Response: produced a structured, bullet‑heavy prompt (price caps, BIO success rates, dilution, competition).

3 . The ​“Cold‑Start” Triplets: What Actually Happened

Armed with the one‑shot prompt, I opened three new ChatGPT tabs:

• Instance A estimated Cidara at $289/​share (pre‑dilution) → $202 after dilute.
• Instance B pegged it at $12/​share.
• Instance C landed at $42/​share post‑dilution.

Why the spread?

• Label breadth odds: A (50 %), B (20 %), C (50 %).
• Senior uptake: A (60 %), B (10 – 25 %), C (25 %).
• Margins: A (65 %), B (60 %), C (40 %).
• Profit horizon: A & C model 20 years; B chopped at 10.

The prompt had price caps and dilution, but didn’t force senior uptake to stay tethered to the 69 % flu‑shot benchmark or make the model keep a generic‑tail cash flow. Result: three very different NPVs — none ​“wrong,” each a different end of the plausible spectrum.

4 . What We Learned About CD388 — and About ChatGPT

Flu market nuggets surfaced along the way

• Seniors already line up for a jab that’s half as effective; a once‑a‑season antiviral with 70 % protection could legitimately command $250 – 300 per dose.
• At 60 % senior penetration and $250 net, CD388 would pull in $15 – 18 B U.S. revenue even after Medicare slashes price in Year 14.
• Pandemic‑insurance value (averting a 1918‑like event) alone could justify higher early‑life pricing.
• Manufacturing a 450 mg drug‑Fc conjugate for tens of millions will test global bioreactor and fill‑finish capacity — an execution risk.

AI insights

• Default behaviour is to comply with the prompt, filling blank cells with plausible but often extreme numbers.
• A single prompt can’t capture every nuance: the model must guess slot‑count, uptake friction, exact dilution math.
• Iterative dialogue matters; each follow‑up acts like an analyst’s red‑pen on a junior associate’s draft.
• User domain knowledge steers the ship: knowing the CDC flu‑shot rate or BIO success tables lets you detect when the model is off the rails.

5 . Take‑home for Non‑Biotech Readers

1. AI is great at the grunt work—pulling analog price comps, crunching DCFs, translating Phase‑2 data into cost‑effectiveness.
2. AI is terrible at choosing the priors—will 10 % or 60 % of seniors get the shot? Will the FDA insist on two seasons of safety data? That’s judgment.
3. You (the human) must interrogate the numbers the way we just did:
   Why only 25 % uptake?
   Where did that margin come from?
   Show me the dilution math.
   Add a mRNA competitor.
   Cut price 20 %.
4. There is no ​“perfect prompt.” The shortest route to a solid answer is a long, curious conversation.

6 . Conclusion: Keep Your Favourite Humans

Could ChatGPT eventually ingest real‑world payer dossiers, KOL interviews, and CMC cost curves on its own? Probably. Will it, in one shot, deliver the same valuation a seasoned biotech analyst retrieves after a week of phone calls? Not yet.

Until then, treat generative AI as a sharp intern who never sleeps—worth its (virtual) weight in PowerPoint slides, but still in need of a mentor with domain expertise and a persistent ability to ask, ​“Are you sure about that number?”

Replication — I did this all again and I got …

Being a good scientist, I replicated my entire process and mentored a whole new Thought Partner AI that ultimately concluded Cidara was worth $500/​share and then offered a prompt that it believed would get new ChatGPTo3 sessions to do an analysis as systematic as it had done with my guidance. I tried that prompt in five separate ChatGPTo3 sessions and got back the following estimates: $43, $43, $113, $150, and $200. In other words, the results are the same as the first time. ChatGPTo3 couldn’t come up with a prompt that would get itself to do an analysis systematically and consistently enough to come up with a result that would agree with it. And if ChatGPTo3 can’t rely on itself to do work to its own standard, then it’s probably not ready for a human to rely on.

Just because no one ChatGPTo3 session should be relied upon for a consistent valuation, that doesn’t mean it can’t be a good thought partner. It made many intelligent observations, sometimes recognizing that biologics don’t go off a patent cliff but erode gradually, that an antiviral – unlike a vaccine – works independently of a person having a functioning immune system, and that the utility of the drug would vary by market segment. 

I didn’t double check its NPV models but they directionally made sense (i.e., when it cited a peak sales estimate, I would find that its risk-adjusted valuation was in the right ballpark based on multiples). 

It usually recognized on its own that one should account for future dilution from financings if the company planned on launching its drug on its own, something that entry-level associates typically have to be taught if they are coming straight out of science. This kind of thought partnership might be useful to a lay person though it’s probably not enough for an investment professional to consider ChatGPTo3 to be a substitute for an analyst with a few years of experience. But I would expect an analyst to get utility from using ChatGPTo3 as an aid to gather references and create an inventory of parameters for them to personally examine and assign values to. 

Which AI got the valuation right? 

If you’re still with me, you might wonder what I think Cidara is worth. As noted above, RA Capital has a sizable position in the company. It was our team that worked with management to license its drug back from Johnson & Johnson. As you might guess, we do think highly of the drug, the data, and the team. As you might also guess, I can’t and won’t give you a target share price. 

But consider that if CD388’s Phase 3 trials prove that it’s as effective as the Phase 2b suggests it is, it will forever change humanity’s relationship with flu. We’ll never have to fear flu pandemics, we won’t have to worry about how well vaccines match this season’s circulating strains, and we will enjoy healthier and longer lives. 

Just consider people over 65, who are highly susceptible to flu infections. There are over 200M people over 65 in the developed nations that contribute the majority of any drug’s revenues. Hospitalizations aren’t cheap anywhere, and even with vaccinations the flu sends many older people to hospitals. A paper from over 20 years ago estimated the total economic burden of seasonal flu to the US alone was $87B/​year, so it doesn’t seem crazy to estimate that the cost today might be $200B/​year for all developed nations. 

So could this be a >$30B drug at peak as my Thought Partner AI suggested? If it were, and if it cut the impact of flu by 75%, it would still be a bargain for developed nations. 

On top of the seasonal market, CD388 is the kind of drug that governments should want to stockpile to protect against a flu pandemic. Since the nature of pandemic flu is that it might be more or less lethal for people of any age (1918’s Spanish flu was more lethal to young adults in their prime than to the elderly or children), the stockpile would need to cover most of the population. So we’re talking about at least a billion doses just for the wealthy world. While the price point for such a contract would be less than for the seasonal market, even a theoretical bulk price yields astonishing figures. The value of a company being paid even $20 of gross profit per dose for that many doses would be enormous. I have no insight into whether governments will want to stockpile such a drug. (It’s logical but governments aren’t always logical.) But even a 25% chance of winning such contracts is worth billions. 

So is my job safe? (And some implications for how companies write press releases.)

So without some guidance, ChatGPTo3 left to its own ​“instincts” really did a terrible job of appreciating even a fraction of the value that I see in such a drug. AI for biotech company valuations is not yet as good as our most junior analyst. Still, we’ll keep coming back to it to test its wits from time to time.

But this exercise does raise new questions for all companies putting out data or any news. What could you be including in your press releases to ensure that AI has enough context and guidance to not grossly undervalue your accomplishment? Because increasingly, press releases won’t only be read by humans, but also by AIs that will tell humans what the press releases might mean. And since AI can clearly reach very different answers given the same limited information, the question is whether providing a lot more information might inform and standardize an AI’s analysis. 

So maybe the Cidara press release could have provided a lot of additional context on the potential value of its drug candidate from the standpoint of productivity, caregiver burden, and hospitalization costs in the US and other markets. It could have spoken to the concept of pandemic preparedness. Maybe it should have included a link to the company’s corporate deck and instructions to study it alongside the details included in the press release. 

There’s room in the market for a PR firm that specializes in drafting press releases that are intended for AI summarization and analysis. I look forward to being pitched by such firms to help all our companies make sure their value propositions are not overlooked. Or maybe my team will build an agent that does this automatically and put it out for all companies to use. The power of AI is crazy that way… what used to be a standalone business potentially becomes just a feature of some other business.

Meanwhile, it’s likely that in the time it took my (still human, for now) editor to review this draft, argue with me that it’s too long for even a dedicated AI to finish, and prepare it for publication, ChatGPT and other AI agents have improved a bit. The AI agents today can’t quite ​‘get’ the flu – or the value of a flu company – but they’ll probably get there eventually.

Heck, the day that becomes possible, RA Capital will hopefully be the first to launch a low-fee AI-traded biotech ETF (alongside the others we’re ​“working on”). ChatGPTo3 already even came up with a couple potential names for something like that … it’s itching to get started. Buyer beware.

AUTOCLAVE – Autonomous Life Science Trading and Value Engine

“Sealed tight, super sterile, and kills 99.9% of human investing error.”

PETRI – Predictive Equity Trading Robot for Innovation

“Grows your money like E. coli at room temp. Except cleaner.”

Disclaimer: RA Capital owns a position in Cidara. Nothing here should be taken as a recommendation to invest in Cidara or any other security. Nothing written by ChatGPT should be taken as a verified fact and could be wrong (in fact, showing how inconsistent it is was the point of this article). Neither this article nor any of the prompts referenced here and in the linked PDF transcripts of the author’s discussions with ChatGPT should be taken as fact. Readers should verify all information they read anywhere, including AI-generated content.