When will the panic alarm sound for AI

Yeah it’s fr scary to think about and given enough time will certainly happen.

This next bit is me parroting my gf who’s an AI researcher in a FANG sub (so take it with a grain of salt)

Basically the current LLM model cannot “learn” even maxing out the quality data and compute power (which is where its gains have been coming from the past 3 years) it cannot reach this computer god level.

Not saying other stuff isn’t on the rise and unknown but current models are likely never going to get there without serious rebuilding of foundation.

Scary stuff though certainly. Especially with so much confidence from AI peeps about it reaching computer Jesus by 2027

Personally I hope it doesn’t come to fruition without intense government oversight because as you said there’s a huge fraction of the country (and globe) immediately out of a job (honestly long before this point but I digress) which starts falling domino’s

New AI tool just came out that can build complex financial models for you. Its better than most analysts already.

We’re 1-2 analyst classes away from the beginning of significantly lower staffing levels for juniors.

Ai should be the most talked about topic in all of politics (instead we talk about utter nonsense instead). The Geoffrey Hinton interview I saw a few weeks ago was one of the scariest things I’ve seen. Also, he left Google and Ilya Sutskever left Open Ai because of the lack of focus on actually safely releasing these products. We have the most powerful technology in the history of the world in the hands of businessmen who disguise themselves as world-saving geniuses.

This will certainly impact junior recruiting a few years down the line. Jobs aren't going away, but the output from each employee will be significantly greater than we can imagine. CIMs, pitches, models, buyers lists, etc.; These are all things that can easily be created by AI today. 

I think we'll see the first wreckage at Big4 PE service groups doing TAMs and CDD groups in the next recruiting cycle, and it will only get worse. This is not just for banking, it goes for all knowledge workers. 

Junior roles won't go away, but classes will be half of what they are and the traditional pyramid team structure will turn into a line structure. Leading AI, error detection, and knowing how to get valuable output from agentic software will become the most important junior skill. Analysts, together with their associate, will be editors/superiors for the agents controlling output. Team could be 1-2 analysts, 1 aso, 1 VP, 1 MD and get the same output as a 15-20 person group of today. I think this will lead to better pay for the few remaining analysts, as retention becomes more valuable to the bank in order not to exhaust their entire junior pool (future dealmakers). 

AGI will not emerge from LLMs. 

Will actual AI transform our world sooner than any of us expect? Probably. Will the generative slop machine (ChatGPT, Grok, whatever) be the one to do it? No. 

To OP’s point, I think a lot of random “bullshit” middle-manager jobs are gone in the next 10 years. You know that guy in operations or woman in marketing that each get paid $150k and you have no idea what they do to add value? Yea, those are gonna be gone. 

What this really impacts are kids today and how they should be guided career wise. The idea of a middle-class career will change as many college degrees are going to rapidly lose value and trade skills will rapidly go up in value. And there simply won’t be as much of a need for so many random white collar jobs, and those that are available are going to be hyper competitive, even vs today’s standards. You can already see this dynamic playing out in coastal cities where trade workers can charge through the roof for services (if you can even find someone to take the job, yes, it’s that bad) because autistic SWEs and consultants (of which we’ll need much fewer in 10 years) have no idea how to unclog a drain or repair some drywall. Weird times. 

I've tried using AI for an SRT cash flow model and it was shit

So I don't know when this revolution is coming

I think wealth inequality widening is already growing and AI is going to increase that.  And catching up will be harder and harder.  Public companies will cut costs (labor).  Those with assets see the growth of the cost cutting and market positioning.  Those relying just on income from their labor will have a harder time keeping up.

It is crazy to think that 5 years ago, we were widely adopting virtual work during COVID and now this AI.  Another 5 years, and the AI agents will be taking over. 

Learn AI.  There’s more to it than ChatGPT.

There is a lot of opportunity.  From changes in AI search optimization, AI agent implementation for SME’s (especially localized services), to new service jobs suddenly becoming economically feasible (new industries/professions being created now), as a college student you need to be part of the wave or endure a scarcity mindset in the old ways (smaller finance analyst classes). 

There are lessons from my real estate career that I see as patterns.  One is competition is going to be brutal and global, but strategic location creates mini-monopolies.  When there is a glut of supply, getting noticed will attract a premium.  So, I buy premium internet domain names.  It’s the only IP I know of that will last longer than an incremental performance advantage that will be filled by just another AI powered company within months.  That’s a prediction.  

If you have any other suggestions, please let me know.  Or if you disagree with me about domain names being relevant, then help me stop spending money on them (I’ve seen the prices go up a lot).

There is advantage where being local is better, and human touch is a differentiator (on-site management vs off-site).  Focus on building brands locally.  AI for locals.

Well, what if we went into finance because we like it and I don’t want to be an AI implementer?  This is the crappy thing in that I don’t like doing things, I don’t like doing.  So, you need to change your mindset and think about what [your area of] finance will be 20-30 years from now - and want to be a pioneer of that (or get out and find a new industry - currently in place or soon will be).  That is the only way to get motivated to learn these tools as a first mover in your field.

Lastly, I’m very concerned for my kids’ future.  They are elementary school aged.  Part of me thinks I need to start businesses for them to take over, because they won’t be able to get jobs.  Another part of me thinks that’s fine, no job, they can come live with me forever and economically family (human capital) becomes more important and valued.  And then another part of me thinks there are limitless ideas to pursue and dream up that lower barriers to super computing, unlimited energy, and boundless imagination would mean we get to work on things we like doing more.  Maybe we won’t be rich rich, but we will be ok.  

Just like the internet created a long tail of information, distribution and online communities, AI will create a long tail of work with less mundane tasks.  But this will require us to become more entrepreneurial and creative, because if the traditional work is scarce, the free market of your value is exposed and very variable.  You will be a free agent.  The transition to this new normal is going to be rough for many who only know or want stability, but humans are adaptable.  Freedom from mundane task sounds pretty good.  More value placed on human capital is good.  Using your imagination is fun. 
 

The uncanny valley will always exist. It is so easy to spot the drivel spit out by AI. Maybe software engineers could engineer themselves out of jobs but AI will fundamentally be a tard and all the anxiety around it is born from Hollywood. If anything it will help our productivity as we experience a baby bust and each generation gets “lazier”. TLDR I wouldn’t worry about AI.

I was a junior partner at a $100M venture capital fund, I now run my own software engineering training and outsourcing company, and I have a computer engineering degree, so here are my two cents.

TL;DR: 

a) AI is disrupting junior roles yes, but world is not ending and new jobs will appear or old ones will morph;

b) Our junior software developers are heavily hit, but we're adapting by turning them into AI engineers and doubling down on critical thinking skills;

c) AGI is not happening with LLMs, the Silicon Valley tech bros keep hyping it just to justify inflated valuations.

 Longer Answer :

An LLM is basically a parrot on steroids that does next-word prediction. If the field you're training it on has a lot of repeating patterns in its data, the resulting AI model will be really good at giving the answers you need. 

This is why LLMs works so well in coding, because a large part of any codebase is made up of common functionalities (login button, write, delete, update, etc...) and thus same patterns of code are found across all codebases and will be reused for new ones. Human languages and their sentence structures are also not that varied, which is why LLMs reproduce human writing well.

Coding work at the junior level was mostly about copy and pasting code - there's similarly a lot of donkey work in junior finance - and this is the part that LLMs automate quite well. Instead of opening 20 tabs in the browser and copying from there, the LLM fills the content for you from its trained model.

Obviously that doesn't mean I'm super happy about it, because we went from not being able to handle client requests for our junior engineers from 10 countries, to having to scale back our growth expectations drastically. My company is definitely being disrupted by the AI wave.

So how are we adapting? Well we are aggressively updating our pro training programs and injecting AI, obviously.

What is more interesting is that we are also doubling down on critical thinking. Whereas juniors before could get hired and coast on copy pasting code, that doesn't work anymore, and you have to be able to do two things now:
1) Be familiar with the latest AI building tools, so that you can prototype a proof of concept (POC) for your manager or client very quickly (we're talking building an app prototype in 3 days instead of 3 weeks)

The above means you can still create value for the client. Yes the non-tech client can now vibe code the app themselves, but this has its limitations; you still need someone who can look under the hood.

2) You also have to have the deep analytical skills, the critical thinking ability of a software engineer to trace code written by AI in order to fine tune it and clean it up. The AI tool can build a POC quickly, but if the client likes it you'll definitely need to  tweak the code, secure it, maintain it, scale the database, etc...

So that's basically how we're solving the disruption in the coding world for juniors. We raise the level of critical thinking needed as a software engineer, while training on the latest AI tool.

Is there a similar adaptation possible for IB? We'll see.

So that answers a) and b) in my TL;DR.

As for point c (AGI or superintelligence is coming and we will all be out of jobs), I don't buy it. LLMs don't reason, they merely repeat. You need a completely different paradigm, a new AI approach different from LLMs, and I don't know of any such so far.

Finally, on a bit of a lighter note, some advice for would-be founders: As an ex-VC, I have been through multiple tech trends, such as Big Data/ML, blockchain, and now LLMs. We know that founders make sure to inject the current hot keywords into their investment decks in order to show VCs that they are at the cutting edge.

Many of those founders' product will have nothing to do with AI/blockchain/etc..., but they still feed the investors a bit of bs in order to impress them :)

What I advise is not to do this unless you're ready to answer "how are you really using LLMs in your product and how is adding them creating value". The good VCs can smell a fake pitch, because we see so many.

Hope the above helps! If anyone has venture capital or technology-related questions, hit me up.

I'll throw my hat into the ring as a software engineer that actually understands how AI/ML works, especially LLMs. So buckle up for some hard hitting reality. I'm also developing a new product that utilizes AI so I'm keenly aware of the cost of using it.

To answer your first question, we are no where near AGI. GPTs are basically really good prediction machines using the same math and stats as a basic neural network, except that it's trained on billions of parameters of data. It doesn't actually "think" in the sense of having emotions or feelings, but more so, here is the task I am given to accomplish, here are the constraints, here is the context, now I must plot a course to solve said task. 

If you actually use AI for data entry, you'll actually find that it won't replace people, rather it will just speed up data entry tasks and you still need people to oversee the AI, albeit maybe less of them. I've tried to use Chat GPT multiple times for web scraping and extracting data and it shit the bed every single time. However, for my new startup which solves a personal pain point, Chat GPT as well as any other GPT is incredibly good at summarizing and extracting key information (which is what it was built for in the first place). 

Gen AI is incredibly energy inefficient and still incredibly expensive. There is a reason they are building out the data center in Texas, but even then we won't know how long it'll take to actually finish construction and how much power demand it will need. Of course as time goes on, with much better Nvidia chips for training and computation, the cost of compute will hopefully go down, but right now, it's still incredibly inefficient and you can't see massive adoption if the cost is too high and unjustifiable. So we'll definitely be using AI for a lot of tiny misc tasks, but it won't completely take everything over because 1. A monkey can do a better job at it 2. It's too expensive especially if you look at cloud compute costs to operate it. 

AI does not replace manual labor. Not in the slightest. AI is also dumb as shit. Again, it's only good for replacing grunt work tasks or for tedious digital grunt work. It's sole purpose is to automate the redundant work so that you don't have to think about it. There are some excel wrappers out now, but again, it can do perhaps 80-90% of the work, which is amazing, but for the more niche and complex stuff, it will require a person to go in and work on the model. A caveat though, you forget that banking is not just financial models and pitch decks. At the junior level it's you just building out slides, models, and scheduling meetings. The real money is at the Director + level where you are working in a pure sales role and expected to solve your clients problems (i.e. we need money help us find investors or we need to sell of this port co of ours, we're too busy please market this for us, or we're looking to merge our companies but the work to go through all of our assets is too much so we need outside help with making sense of this and figuring out what is a fair deal) Notice how in all of these instances, an AI can help speed things up, but it will not replace humans as sales is fundamentally human. If anything AI might end up finally getting rid of the pay your dues culture and now force banks to realize we can speed up our sales training pipeline, minting new MDs. 

I wouldn't worry too much about AI, instead embrace it, as it's the second closet thing we have to the new internet for the time being. 

Before that, you had to go to experts to do some random grunt things and you had to pay them. Now, you will be able to do lots of things on your own without having to spend any more for outside services. So, everyone just maintains its same level of wealth and more people will be redirected towards work that requires more brainpower or manual labor. The fact that you're not spending $1000 on a lawyer to draft a contract will be spent probably on your vacation somewhere else, which will accrue to that hotel. This is great for efficiency and for quality of life, as we prevent lots of money going to "necessary evils", and instead we can allocate that amount saved to relative good things that improve our quality of life. That lawyer, on its own, might now need to work at that hotel to give you a good experience. He could do better, maybe be a relationship manager in a private bank or doing some Government work, so it's just an economic shift. This might raise some concerns related to the inequality gap, but instead of creating a "fictional" white collar class who relied on such tasks (e.g., lawyers, bookkeepers, translators, etc.), you'll have more wealth going towards those big entrepreneurs and a more balanced working class without the same discrepancies we had until now about white collar jobs and blue collar jobs - which means that a rational actor should be indifferent towards a white or blue collar job (albeit a case could be made that white collar jobs, being more prone to competition by savvier users of AI might be riskier vs. specializing in a craft and get recurrent business in regional markets because of your seniority in the field). Additionally, this not an issue if in the light of this inequality, the taxes can be increased on entrepreneurs to avoid extremely high concentrations of wealth due to the shit of what before was wealth accrued by white collar jobs into their hands, and instead, reduce the tax burden of those classes which might not make the same amount of wealth on a total basis as they did before those models. Again, for the juniors and those aiming for those jobs that are at risk of AI, any fear related to AI is irrational. The raise of white collar jobs is something fairly new which appeared in the late 60s, until then, almost +80% of the workforce was in blue collar jobs so this is indifferent for the economy and for the individual who is looking to join the workforce - the case could be made that it's better as a whole because now access to knowledge or intellectual-oriented tasks will be cheaper due to an increase of efficiency or, alternatively, more entrants becoming more sophisticated in the field (meaning, a higher quality of services at the same price as before)

For what it is worth, I tried out the AI tool that has been going viral on linkedin, etc. - asked it to fill out a very straightforward three statement model from a very well signposted template and it failed pretty spectacularly. Not to say that it won't be better in the future, but I doubt the veracity of their claim that their AI is already better than analysts...

I guarantee there will be a gigantic recession based on a combination of the private capital super bubble bursting along with all asset values resetting finally to reflect the higher for longer interest rate environment due to the runaway national debt, before AI is the reason broker dealers or banks cut front line people.

Also the global supply chains for electronics and electricity itself will break down before AI is a panacea. We will need to go back to making things here jobs and much less email jobs.

I work in healthcare consulting and had this conversation with a partner, he thinks there is some bullshit but some of the clinical research capabilities are incredible from what I'm seeing in the industry. Some of the clinical support AI is also very cool. I can see younger clinicians using them.

Medical scribe / documentation AI is a hit or miss. But I've also seen a medical doctor retiring soon love AI creating work notes for patients.

But for the most part for the healthcare administrative side, I'm very skeptical. A lot of them have a ton of oursourced employees overseas and say they're AI (agentic AI functionality). Sounds sketchy but some of them are backed by reputable investors (General Catalyst is throwing money at healthcare AI, I've seen videos of the GC CEO talking about why healthcare and he doesn't seem that knowledgeable at all - he just wants to scale his platform to become an incumbent). Despite the skepticism, if I eventually exit in the next year or so - if this hype train still exists, these are the companies I would target.

I made a thread about Claude for Financial Services but it got no traction. Claude Code is the market-leading product that is "accelerating software development". Take a look at how the product works to get an idea of the workflow. It has changed the software development workflow entirely. I don't work in financial services but from afar, I do not see why it would not work for Financial Services' work products.

I think @QuiltEmerson has the most rational take here. Gradually, but most assuredly, AI will impact professional knowledge workers.

AI isn't the primary diease, it is a symptom of a society that has pretty much given up on forward progress. 

When you start realizing that AI is a useful tool and not a magical leprechaun that does all your work for you, it's a lot less scary.

CEOs and tech bros sell it as a magical leprechaun that does any kind of work with a click of a button. It's not that now and I don't think it will get anywhere close to there.