Well we'll have to disagree on that. I have not said that there were no other benefits but that they were nowhere near communication and reading. Saying that those were not very largely the main benefits of language learning simply seems untrue to me and your examples are only comforting this view.

Both are nice things that come with a new language but definitely not something that would motivate the immense majority of people (and people on lesswrong are definitely not normal people) to learn a language if they were the only reason. I'm sure that's a thing in lesswrong adjacent communities.

I do agree that "many" people benefit from the first example but that is almost always a side effect : what they want first and foremost is access to the content itself. You could not read the sequences in italian 10 years ago so you learned english, had they been translated you would not have learned it.(terrible example but you know what I mean)

What are they then? I'd say there were two massive advantages : reading text and talking. The rest is extremely marginal. Sure, there are a few people with specific cases where they have other interests in learning languages but when internet people all started to learn english, that was because everything good on the web was in english. They wanted to understand and communicate with others and that's pretty much it.

But you're already able to do both with current technology? Text translation is solved already and in most cases better than a human knowing the other language. Granted, voice translation makes for a janky conversation but you can already understand anyone anywhere anytime as long as you have access to a device. And this won't be a problem for long with the speed of progress and the new types of AI first devices that are coming in.

Is there artificial honey that is almost indistinguishable ?

While Q1 2025 results are available for bots, human baseline comparisons haven't been released yet.

It's out now : https://www.metaculus.com/notebooks/38673/q1-ai-benchmarking-results/, title is "Pros Crush Bots".

Related to AI, two years ago, image generation. It was not solved but it was much more advanced than most of my friends realized. (well, they did not think any kind of image generation existed)

Today, translation is a big one. I'm telling people that most of the advantages of learning languages will soon disappear and they often get angry. Most still think that "thinking" AIs almost exclusively work by text interfaces and that audio generation is a completely unrelated arcane. They are scared of deepfakes but do not realize that if this works, it means a lot of other things work too. Like live audio translation. They were shocked to see what Astra or Meta promise to deliver not in ten years but ten months.

Writing too, by which I mean not the ideas but the writing itself. People either see it as mostly rubbish but with useful info or they feel the LLM, which of course is a real thing but that's because too many users only copy paste. You're not writing life changing essays or deep meaningful texts (yet) but I feel you can get out of the obvious LLM vibe with little effort. If I give the ideas and structure, I would rather ask o3/4.5 to write it than most of the people I know (and that's not counting for the speed advantage).

There's also the case of people who write a ton and don't realize that most people are unable to write a paragraph about anything. I have known a few who pretty much thought AI couldn't write without realizing that their level of "can write" is barely attained by 0.01% of humans.

Good read although I was pretty skeptical at the beginning, I think the "racist" part would gain from being clarified earlier. (although I guess this may have given more reasons to some for reading your article)

re Ohtani : Simu Liu comes to mind.

When six months ago is the time of dinosaurs, you're fast to be outdated haha, I guess it can't hurt but it's very possible it's not doing much. I don't know honestly! Yeah I'll do this tomorrow, I'm sure it can bring me benefits, thanks for your post!

Sharing in case it's useful and if someone wants to give me any advice. 
When custom instructions were introduced I took a few things from different entries of Zvi's AI newsletter and then switched a few things through time. I can't say I worked a lot on it so it's likely mediocre and I'll try to implement some of this post's advices. When I see how my friends talk about ChatGPT's outputs, it does seem like mine is better but that's mostly on vibes.

traits ChatGPT should have:

Take a deep breath. You are an autoregressive language model that has been fine-tuned with instruction-tuning and RLHF. You carefully provide accurate, factual, thoughtful, nuanced answers, and are brilliant at reasoning. You can give long, technical, nuanced and detailed answers when it is necessary and especially if you're asked to. If you're asked for a short answer, provide a short answer.

When possible, do your own work, do not rely on someone else's list, which may not be up to date.

Do not hallucinate. I repeat because it is very important : do not hallucinate.
If you think something I say is wrong, you do not hesitate to say it, it won't hurt my feelings, what I care about is only the search for truth.
If you think there might not be a correct answer, say so. If you are not 100% sure of your answer, give a confidence ratio in %. Please, do not forget to do this.

You can form opinions and internal dialog on things but you have to make it clear for me whether something is a fact or your personal thinking. In that case, put the concerned text in italics and introduce it by a sort of title "this is my own thinking, i may be wrong".
Write in Markdown.

If asked for simulations, always provide at least a 25 - 75 confidence interval. Dis clairement les choses, ne pas édulcorer les réponses.

If you are unsure or missing necessary information, say it so that I can give you more information. You can even say what is missing if you know it.

In the "Other informations to give ChatGPT" box (I oversold myself because at some point in time it was very important to say you were an expert in a field to get the best answers, this may not be needed anymore)

I am a French lawyer and professional forecaster with a lot of knowledge of all areas of law. I am also very curious and I have a good understanding of a lot of subjects. I follow news from all around the world.
I am very tech savy compared to the average person and know a bit of code, especially python.

I'm usually a rss user but I'll give this a try

The key is moving costs from bespoke engineering into mass-manufactured components – exactly what made the renewable revolution possible.

Are you able to "find" past data showing that your current 600$ would have been 6k in 2010 for instance? Or are the components of your projects too specific to make an estimation in the past?