Hi everyone, I’m thrilled to announce that The Algorithmic Bridge has reached 10,000 subscribers!!
Before we go into the AMA, I want to thank you all for being part of this tiny internet corner that is TAB. In special, those of you who make it financially viable by contributing with a monthly/yearly subscription. You’re the reason I can focus almost exclusively on distilling into (hopefully) understandable essays how AI is shaping our increasingly complex world. I also deeply appreciate those of you who week after week engage in thoughtful debate in the comments (even if sometimes I take my time to respond!) and who like and share these articles with your family, friends, and colleagues.
I had planned to do an AMA for 10K but didn’t expect to get here so soon. A couple of Twitter threads went semi-viral and brought many of you here (Hi!). In case some of you didn’t see them and want to:
So let’s get to today’s thread. As you know, you can use this AMA to get out of your chest those questions you want me to answer that I don't cover in the articles because they're too peripheral or have nothing to do with TAB itself. The “anything” in AMA is almost ;) absolute. Be as specific as possible, it really helps me frame my answers better! Some ideas for your questions:
AI: current trends, predictions, models (ChatGPT, GPT-4, Bing, Bard, etc.), companies, famous AI people…
Writing: Substack, my research and writing process, future plans for the newsletter, things you’d like me to change or cover more deeply…
Things peripheral to AI: society, culture, science, business, politics…
And anything else as long as you do it respectfully.
I haven't seen much discussion about the general public's reaction to AI. Currently AI awareness still seems restricted to a very small subgroup of people.
Whenever I talk to others about AI, even many tech workers (!!!), 95% seem utterly unenthusiastic and never demonstrate a single ounce of curiosity about the AI's capabilities or what it may bring. Managers and executives, the more senior they are, actually demonstrate much more interest into AI.
This just feels utterly worrying. How can people see a tsunami coming and don't care? It feels like these people, are just going to put their heads in the sand. And when GPT5 or GPT6 rolls over, they will confront these AIs with 0 preparation or time to cope, and cause mass social turmoil/roits.
Any plans to write more on the public reaction to AI? And how the AI community should communicate to those who are not in the know.
My hearty congratulations on this splendid achievement, Alberto!
Here's my question: how have you integrated the use of AI-based applications and products into your day to day routine? Either for personal requirements or work outcomes. Curious to know!
Question about What do you think Microsoft and OpenAi’s strategies behind launching both Bing and ChatGPT + browsing. Because to me the with browsing I feel I don’t need use bing anymore. I’m really curious about it
How long do you think it will take for more commercial uses of generative AI to take hold (i.e., big companies incorporating it into everyday processes)?
I would love to incorporate AI as a kind of “research assistant,” but so far it’s been disappointing at least with ChatGPT. Maybe I am using it wrong, but the responses are so general, often unreliable, and unimaginatively repetitive. Not very useful for the social sciences or as a writer more generally. Any advice? Or should I just wait until the tools improve to even try incorporating this?
An all-purpose researching bot is really what I have been dreaming of - given I have so many notes and ideas, but not enough time to parse through the web searching for source material. This would really be that 0 to 1 moment for nonfiction writers. So far though, it’s not working out for me.
A colleague who is a paid subscriber gifted me a subscription to sharpen my musings on AI. Say, what's your stance on the latest efforts to regulate AI? I am not even sure which part of the world you are inhabiting. And, out of curiosity, what drove you to dedicating your writing and thinking to AI?
So cool to see you hit this milestone - congratulations Alberto!
As a curious fellow Substacker, how did you see your subscriber growth curve evolve? Was it a gradual / linear growth or did you see some occasional big jumps, followed by relative plateaus? What in your experience were the types of posts that tended to bring in the most subscribers? Were they detailed opinion pieces that you're famous for, or were they more practical posts with advice, tips, and insights? Do particular channels seem to do better than others (e.g. organic vs social)?
Any interesting observations you may have made about how Substack has worked for you would be awesome to hear about!
A conspiracy theory that I find amusing is that OpenAI keeps releasing so many groundbreaking products because they have already built GPT-5 and its doing all of the coding for them!
But I think there is something about having foundational models and being able to layer on top of them to accelerate the pace of progress. Do you have any insights into why OpenAI is able to ship so much, so fast?
Hi Alberto, thanks :) also for answering all three of my questions. Your name reminded me of John Romero, who's from the US, now living in Ireland.
We are on the same page: Policymakers should make the effort to understand how AI systems work. And they should understand them before they regulate. The dilemma is the acceleration of AI progress, driving policymakers into rushing decisions (because regulation is obviously needed – and lagging). Companies with deep pockets have other means to fulfil compliance requirements. Our European economy is small-scale/ scattered and will be affected in different ways. A difficult task for regulators. Self-regulation of platforms doesn't work, as we have seen.
Your polymath approach is highly appreciated, this is why I am here. As a society, as human beings, we need more meta-reflection like yours to fully grasp and actively shape the ongoing disruption. We have some things in common, perhaps. My background is in culture and history. I assume yours is in tech?
Looking forward to forthcoming episodes of The Algorithmic Bridge! ✍
I feel like there is a wave of generative AI start-ups that kind of resembles what happened with the web3 (especially NFTs), but I'm unsure of how well these companies will survive. What do you feel are the best jobs (and companies, or types of company) for business students looking to work on building products that integrate generative AI, and that are likely to be around for a while ?
I'm (day)dreaming of a path to AGI where LLMs would be used on COMPLETE multi-modalities i.e. not only text or images but according to our own human modalities, i.e. our 5 senses + proprioception.
This would bridge the gap from (autistic) pure language "grasp" of "reality" and our grounded or embodied cognition. Then when the computer would encounter words like "I grasp", "I'm hearing you" "I can smell it" etc. it would have a better "grasp" of a commensurable "reality" with us.
Now there is also progress to be made on causality identification and (human like) reasoning.
Note also that super intelligence wouldn't be limited to our 5 senses but could embrace all the electromagnetic spectrum to benefit from other senses....
We are still so far away from mapping the brain various channels of communication (more than 40 neurotransmitters, 3 or more concurrent spiking waves activation schemes, contribution to (continuous) learning and memory of the dendritic trees formed by all the synapses, etc) that I'm often wondering how we could have already done so far with our poor brain ersatzes that are current ANNs)
Anyway, keep up the good work here, I'm really enjoying this newsletter, we are witnessing an interesting time indeed.
Congratulations and thanks Alberto. It is nice to have a forum with other people excited about generative AI! There are so many tools out there that it is somewhat dizzying. The newest one that I have discovered is Lex.page that now incorporates GPT-4. I signed up and am on the waiting list. It is a word processor with AI incorporated that performs like a co-writer. I think that Microsoft is going to do the same with their co-pilot for Word. Let’s see what opens up first…Cheers
I much enjoyed your article on the AlphaCode paper. My impression after reading the AlphaCode paper (and the appendix) is that the "slow positives" have not been dealt with (i.e. were ignored in the evaluation, I could be wrong but it seems the focus was on reducing wrong answers). Slow positives are defined in that paper as solutions that are too slow, with the wrong (= inefficient) O-classification etc. CodeForces does not include a condition or ranking on speed of solutions in the rules. I wonder if you have insight on what the results would be in case efficiency would be a factor? My impression from the paper is that the results would not be near that of a typical programmer. For a system to be competitive with programmers one needs the solutions not only to be correct but (reasonably) fast as well. My understanding of current systems is that they cannot handle the building of efficient code. In their current incarnation, they seem to be restricted to supportive roles for the programmer. Even that role is necessarily limited, as the suggestion of inefficient solutions typically means a lot of thought needs to be put into improving the suggested solution. What are your thoughts on this aspect? I.e. the capacity of current systems to build efficient code?
So far it seems to me that the application that *most* people are actually using these programs for in the real world right now is to generate first drafts of texts dealing with a huge range of subjects. So it looks like the difference this technology will make for the culture as a whole will have very little to do with productivity as classically defined, but will lead to an incremental but real increase in the intelligence of everything everybody writes. (Not that this is a small change.) Outside of a few niche cases in applied graphics, porn being perhaps the one with the largest number of jobs and money at risk, not many jobs are going to be lost or even changed drastically. What am I missing?
Another business question: One of the strengths of TAB is your engagement with your audience. As TAB races beyond 10,000 to the big time, you're likely to become overwhelmed at some point. Have any plans for that? An assistant perhaps?
Whoa! 10,000! You are the man Alberto. Impressive. If you had the time you could probably do well with a "how to succeed on Substack" blog.
I doubt this is a good plan for you, but if you don't already know, there is another Substack-like blogging network called Ghost which has flat fee pricing instead of them taking a percentage like Substack. The percentage take formula may become more of an issue for you as you continue to skyrocket your audience numbers.
I just spent a couple hours checking out Ghost and have a review hitting my site tomorrow morning. Overall, Ghost looks pretty good, though it's a far smaller network than Substack. I have no intention of moving off Substack myself, but I'm happy to have a better idea where I might move if the need arose.
As for AI questions, Mr. Predictable Boomer Doomer still hasn't heard a satisfying answer to this question on any AI blog:
QUESTION: What are the compelling benefits of AI that justify taking on more risk at at time when we already face so many?
I haven't seen much discussion about the general public's reaction to AI. Currently AI awareness still seems restricted to a very small subgroup of people.
Whenever I talk to others about AI, even many tech workers (!!!), 95% seem utterly unenthusiastic and never demonstrate a single ounce of curiosity about the AI's capabilities or what it may bring. Managers and executives, the more senior they are, actually demonstrate much more interest into AI.
This just feels utterly worrying. How can people see a tsunami coming and don't care? It feels like these people, are just going to put their heads in the sand. And when GPT5 or GPT6 rolls over, they will confront these AIs with 0 preparation or time to cope, and cause mass social turmoil/roits.
Any plans to write more on the public reaction to AI? And how the AI community should communicate to those who are not in the know.
What are your thoughts on the "Sparks of Consciousness" comments?
My hearty congratulations on this splendid achievement, Alberto!
Here's my question: how have you integrated the use of AI-based applications and products into your day to day routine? Either for personal requirements or work outcomes. Curious to know!
Question about What do you think Microsoft and OpenAi’s strategies behind launching both Bing and ChatGPT + browsing. Because to me the with browsing I feel I don’t need use bing anymore. I’m really curious about it
How long do you think it will take for more commercial uses of generative AI to take hold (i.e., big companies incorporating it into everyday processes)?
I would love to incorporate AI as a kind of “research assistant,” but so far it’s been disappointing at least with ChatGPT. Maybe I am using it wrong, but the responses are so general, often unreliable, and unimaginatively repetitive. Not very useful for the social sciences or as a writer more generally. Any advice? Or should I just wait until the tools improve to even try incorporating this?
An all-purpose researching bot is really what I have been dreaming of - given I have so many notes and ideas, but not enough time to parse through the web searching for source material. This would really be that 0 to 1 moment for nonfiction writers. So far though, it’s not working out for me.
A colleague who is a paid subscriber gifted me a subscription to sharpen my musings on AI. Say, what's your stance on the latest efforts to regulate AI? I am not even sure which part of the world you are inhabiting. And, out of curiosity, what drove you to dedicating your writing and thinking to AI?
So cool to see you hit this milestone - congratulations Alberto!
As a curious fellow Substacker, how did you see your subscriber growth curve evolve? Was it a gradual / linear growth or did you see some occasional big jumps, followed by relative plateaus? What in your experience were the types of posts that tended to bring in the most subscribers? Were they detailed opinion pieces that you're famous for, or were they more practical posts with advice, tips, and insights? Do particular channels seem to do better than others (e.g. organic vs social)?
Any interesting observations you may have made about how Substack has worked for you would be awesome to hear about!
And again - huge congrats! Go out and celebrate!
A conspiracy theory that I find amusing is that OpenAI keeps releasing so many groundbreaking products because they have already built GPT-5 and its doing all of the coding for them!
But I think there is something about having foundational models and being able to layer on top of them to accelerate the pace of progress. Do you have any insights into why OpenAI is able to ship so much, so fast?
Sparks of AGI: looking forward to read your comments in due course on that article. Seems every time I blink there is a fundamental shift (or not?)...
Hi Alberto, thanks :) also for answering all three of my questions. Your name reminded me of John Romero, who's from the US, now living in Ireland.
We are on the same page: Policymakers should make the effort to understand how AI systems work. And they should understand them before they regulate. The dilemma is the acceleration of AI progress, driving policymakers into rushing decisions (because regulation is obviously needed – and lagging). Companies with deep pockets have other means to fulfil compliance requirements. Our European economy is small-scale/ scattered and will be affected in different ways. A difficult task for regulators. Self-regulation of platforms doesn't work, as we have seen.
Your polymath approach is highly appreciated, this is why I am here. As a society, as human beings, we need more meta-reflection like yours to fully grasp and actively shape the ongoing disruption. We have some things in common, perhaps. My background is in culture and history. I assume yours is in tech?
Looking forward to forthcoming episodes of The Algorithmic Bridge! ✍
Hi
Congrats on achieving this milestone!
And thanks for having this AMA as well
There had been news that Bard’s model is only 770m parameters and GPT4 is about 1 trillion. Is there any truth to that in your opinion?
Also, I am curious are there any other interesting or promising AI models on the horizon that’s worth looking at other than transformer-based?
Lastly, what do you think of forward forward algorithm vs backpropagation?
Thanks
I feel like there is a wave of generative AI start-ups that kind of resembles what happened with the web3 (especially NFTs), but I'm unsure of how well these companies will survive. What do you feel are the best jobs (and companies, or types of company) for business students looking to work on building products that integrate generative AI, and that are likely to be around for a while ?
Hi there,
I'm (day)dreaming of a path to AGI where LLMs would be used on COMPLETE multi-modalities i.e. not only text or images but according to our own human modalities, i.e. our 5 senses + proprioception.
This would bridge the gap from (autistic) pure language "grasp" of "reality" and our grounded or embodied cognition. Then when the computer would encounter words like "I grasp", "I'm hearing you" "I can smell it" etc. it would have a better "grasp" of a commensurable "reality" with us.
Now there is also progress to be made on causality identification and (human like) reasoning.
Note also that super intelligence wouldn't be limited to our 5 senses but could embrace all the electromagnetic spectrum to benefit from other senses....
We are still so far away from mapping the brain various channels of communication (more than 40 neurotransmitters, 3 or more concurrent spiking waves activation schemes, contribution to (continuous) learning and memory of the dendritic trees formed by all the synapses, etc) that I'm often wondering how we could have already done so far with our poor brain ersatzes that are current ANNs)
Anyway, keep up the good work here, I'm really enjoying this newsletter, we are witnessing an interesting time indeed.
Congratulations and thanks Alberto. It is nice to have a forum with other people excited about generative AI! There are so many tools out there that it is somewhat dizzying. The newest one that I have discovered is Lex.page that now incorporates GPT-4. I signed up and am on the waiting list. It is a word processor with AI incorporated that performs like a co-writer. I think that Microsoft is going to do the same with their co-pilot for Word. Let’s see what opens up first…Cheers
I much enjoyed your article on the AlphaCode paper. My impression after reading the AlphaCode paper (and the appendix) is that the "slow positives" have not been dealt with (i.e. were ignored in the evaluation, I could be wrong but it seems the focus was on reducing wrong answers). Slow positives are defined in that paper as solutions that are too slow, with the wrong (= inefficient) O-classification etc. CodeForces does not include a condition or ranking on speed of solutions in the rules. I wonder if you have insight on what the results would be in case efficiency would be a factor? My impression from the paper is that the results would not be near that of a typical programmer. For a system to be competitive with programmers one needs the solutions not only to be correct but (reasonably) fast as well. My understanding of current systems is that they cannot handle the building of efficient code. In their current incarnation, they seem to be restricted to supportive roles for the programmer. Even that role is necessarily limited, as the suggestion of inefficient solutions typically means a lot of thought needs to be put into improving the suggested solution. What are your thoughts on this aspect? I.e. the capacity of current systems to build efficient code?
So far it seems to me that the application that *most* people are actually using these programs for in the real world right now is to generate first drafts of texts dealing with a huge range of subjects. So it looks like the difference this technology will make for the culture as a whole will have very little to do with productivity as classically defined, but will lead to an incremental but real increase in the intelligence of everything everybody writes. (Not that this is a small change.) Outside of a few niche cases in applied graphics, porn being perhaps the one with the largest number of jobs and money at risk, not many jobs are going to be lost or even changed drastically. What am I missing?
A large bravo from 🇫🇷
Another business question: One of the strengths of TAB is your engagement with your audience. As TAB races beyond 10,000 to the big time, you're likely to become overwhelmed at some point. Have any plans for that? An assistant perhaps?
Whoa! 10,000! You are the man Alberto. Impressive. If you had the time you could probably do well with a "how to succeed on Substack" blog.
I doubt this is a good plan for you, but if you don't already know, there is another Substack-like blogging network called Ghost which has flat fee pricing instead of them taking a percentage like Substack. The percentage take formula may become more of an issue for you as you continue to skyrocket your audience numbers.
I just spent a couple hours checking out Ghost and have a review hitting my site tomorrow morning. Overall, Ghost looks pretty good, though it's a far smaller network than Substack. I have no intention of moving off Substack myself, but I'm happy to have a better idea where I might move if the need arose.
As for AI questions, Mr. Predictable Boomer Doomer still hasn't heard a satisfying answer to this question on any AI blog:
QUESTION: What are the compelling benefits of AI that justify taking on more risk at at time when we already face so many?