Category: linkedin (Page 3 of 13)

Existential Comics: Turing Test with a dog.

The AI House of Cards

2024-06-20 / Luis

Many of the people pushing today for AI to be used in all organisations, particularly of the ChatGPT/LLM persuasion, were pushing NFTs and, before that, cryptocurrencies. And the blockchain, of course. How many meetings we had in our organisations (including universities) asking us to think of blockchain applications for research projects or teaching? A few.

Before all that, 2012 was the year of the MOOC (Massive open online course), when Coursera, Udacity and edX were going to disrupt and break universities. Everyone had to spend, sorry, invest on developing online courses because of the fear of missing out (FOMO). MOOCs are popular, but did not affect university attendance at the predicted scale and universities didn’t make money with most courses.

Now “prompt engineers” insist that we are missing out the LLM revolution, that our students are falling behind, as we do not allow our students to use ChatGPT in assignments. The same ChatGPT that uses fake citations (hallucinations, my bad) to inexistent journal articles to write the requested introduction. What are we supposed to teach? How do we infuse critical thinking, ethical dilemmas, discussion, etc in a probabilistic parrot trained violating the copyright of half of the internet?

Can there be real experts on a technology that appeared year and a half ago? If I look at what I do for work, how long did it take me to achieve my current level of understanding? Way more than 1.5 years.

You could think of this text as a rant; yes, I am tired of people pushing AI. There are piles of money being burnt by large companies right now, a field full of grifters, huge environmental issues as these things use as much power as a small country, and plenty of other distractions. Everyone wants to make money before the whole thing collapses.

Now, if you want to read someone much more tired than me, a scathing post, then I Will [expletive] Piledrive You If You Mention AI Again is a great piece.

PS. I crossposted this text in LinkedIn and my WordPress blog. Both of them offered AI to write my post. Sigh. Featured Image: Existential Comics 15

Counter of La Picada de Pedro, showing lots of yummy food.

Breeding: simple interfaces, complex strategies

2024-06-14 / Luis

I found this text I wrote 20 years ago(*), part of a discussion document I prepared for a review of the radiata pine breeding strategy. Fixed a couple of typos, but I guess we still need pretty much the same thing. 🤔

“Different breeders value different things or, better put, they emphasise different values when developing breeding strategies. One of the reasons why many breeding programs struggle to achieve results is that they face an extremely complex list of activities, which are almost impossible to complete.

“A knee-jerk reaction from some breeders has been to recur to the KISS principle when developing breeding strategies. Unfortunately, the typical reaction has been “let’s create this dumb down strategy because it is simple to apply”. Bzzz. Wrong answer! What they have often done is to create a glorified “deployment strategy” that has almost no chance of surviving in the long term: that is, short term gain based on long term disappointment.

“Breeders need to realise that what needs to be simple is the _interface_ of the strategy. This means that we need a smooth interaction between the “theoretical animal” and the people that will be implementing it. This does not mean that the strategy is theoretically simple, but that the day-to-day activities are a breeze to complete.

“This type of interface requires the development—either in-house or through contracting the service—of tools that make life easy. For example:

Easy access to predicted breeding values, including desktop and online access. In addition, there needs to be an idea of the reliability of those predicted values if we are going to use them for deployment purposes.
Tools that make easy deciding what to select and which trees should be mated with each other (mate selection and allocation).
Protocols for deployment and tools for keeping control of the availability of genetic material.
Easy management of the interaction between improvement and deployment objectives.

“In summary, breeders need tools for dealing with the huge amount of data created by breeding and deployment activities, so it can be transformed into information.”

(*) Well, 19 years ago, this was 2005, but twenty sounds much better.

Are PhDs a pyramid scheme?

2024-06-06 / Luis

If you are a professor in academia you are supposed to form/train new PhDs. In the past, those new PhDs would go to other universities and then train new PhDs, and their PhD students would go to other universities repeating this process ad infinitum. At some point, 20 years ago, 30 years ago—I don’t really know—we reached PhD saturation in academia. There were way too many PhDs for the number of available positions in universities.

It should be obvious by now that thinking of the package [getting a PhD + job in academia] is close to believing in the Tooth Fairy, Santa Claus or whoever is your favourite imaginary character. This doesn’t mean that you should not study a PhD, but that there should be a clear consideration that you will (most likely) have to work outside academia.

I struggle with this as a supervisor. I cannot in good conscience supervise a project *unless* I can see that the project/topic will set the student with skills to work for industry or some other avenue that is not a university.

In my personal opinion, PhDs for future academics are indeed a pyramid scheme.

Why is this trait I like getting worse in the breeding programme?

2024-06-05 / Luis

The short answer: because the trait you like is not part of the breeding objective and, therefore, has not an economic weight assigned to it. And if it doesn’t have an economic weight it has 0 (zero) economic importance.

A longer answer: in breeding there is a distinction between objective traits (which have an impact on profit), and the selection criteria (variables that are easy and cheap to assess, and that are correlated with the objective traits). They may even happen at different ages. For example, in forestry stem volume and wood stiffness at rotation age (say 25 years) can be objective traits for the production of structural timber. Stem diameter, wood density and standing tree velocity at age 8 can be selection criteria.

Some of the confusion may come from when people like a selection criterion (like wood density) and think the breeding programme is trying to improve that. In this example, we weren’t (at least at that time). We cared about volume and stiffness. Sacrificing levels of some selection criterion while pursuing the objective traits is perfectly fine if I am maximising value. And in a modern breeding programme you are pretty much always looking at value, not at a single trait.

If you find this interesting, you may also like Why did my breeding values go down?

Back of the envelope calculations: pulp mill

2024-05-12 / Luis

Imagine that someone stops you on the street and asks “How many hectares of plantations do we need for a pulp mill that produces 1 million tonnes per year of Eucalyptus pulp in Chile?” They don’t need a highly accurate result but a ballpark figure, the right order of magnitude. A Fermi estimate.

How many assumptions do we need?

We need 4 cubic metres of wood for a metric tonne of pulp (wood density 0.5 ton/m³ and 0.5 pulp yield)
Harvest age 12 years
Productivity 25 m³/year/ha

Using 1. we need 4 m³/ton x 1,000,000 ton = 4,000,000 m³ of wood per year. Using 2. and 3. we see that 1 ha produces 25 m³/year/ha x 12 year = 300 m³/ha.

Therefore we need 4,000,000 m³/year / (300 m³/ha) = 13,333.33 ha/year and because we need the same amount in year 1, 2, …, 12 (Harvest age) and we keep on planting forever, the total is 13,333.33 ha/year x 12 year = 160,000 ha.

If you have been paying attention, you’ll notice that we divide and multiply by the rotation (12 years) so we can simplify the calculation back to:

product conversion (4 m³/ton) x capacity (1,000,000 ton/year) / productivity (25 m³/year/ha) = 160,000 ha.

We know that none of those numbers is perfectly correct, but put together they give us an idea of the magnitude of the problem. We can play with them: change site productivity, conversion rate, add safety margins, etc.

Now let’s say that we read of people complaining because a Chilean company announces a 2.5 million tonnes short fibre mill in Brasil. That would need 160,000 ha x 2.5 = 400,000 ha. Massive. As a comparison, INFOR tells us that the whole Eucalyptus estate in Chile is about 900,000 ha and that’s already used by the existing pulp mills, bioenergy producers, etc.

Just from the resource access point of view, having a pulp mill that size would need increasing the country’s Eucalyptus forest estate by roughly 50%. That gives some context to the speculation about the reasons for the investment in Brasil.