* Innovating: A Doer's Manifesto for Starting from a Hunch, Prototyping Problems, Scaling Up, and Learning to Be Productively Wrong by Luis Perez-Breva
* Fall In Love with the Problem, Not the Solution: A Handbook for Entrepeneurs by Uri Levine
These three books really changed my viewpoint and I've been rereading them every year.
Each time I have been laid off, the opportunity afterwards has been significantly better. Perhaps, I have just been lucky.
After my first lay off, I got a job at Sun Microsystems in 1999. I was able to buy a house. After my second lay off from Sun in 2007, I was able to receive a significant promotion as a director. After my third lay off in 2017, I was able to find a great opportunity at Walmart where I no longer have management responsibilities.
If I hadn't lined up my next job so quickly, I definitely would have started my own company or consulting business. The most important thing is to believe in yourself, stay current, and prepare to ride the next wave in technology. :-)
Small world. That sales job was a sales/system engineer at Sun. My world-class sales rep left. For a while, I handled both jobs at once, and it was easier than working with the ding-dong they hired in her place.
Left for a startup where I had the title of I.T. director, though that soon became the entire I.T. department. (ouch) But such is life. It tends to be maximally weird.
I think that we need a better definition of creativity. I suspect that ChatGPT is merely derivative (like a person who reviews all the ideas out there and attempt to pick the best ones) as opposed to original (breaking the conventions typically by a person with a unique viewpoint). This begs the question of what is the definition of creativity and how can we be sure that human creativity is not also derivative. I am scratching my head on this one. How can we define creativity so that it clear that human beings can be original without being derivative?
If you are asking me. Of course, I've spent many hours on it. I have not seen anything that I would consider original. I am very impressed by the quality of responses and how well context impacts the content. It's seems most amazing at summarization and categorization.
I am very surprised by the results of "Let's play Dungeon and Dragons where you are the Dungeon Master" or Write new song lyrics or Complete the following poem or even write the following essay or chapter of a book.
I am surprised by the quality of the response in terms of flow. It sounds very much like a college undergraduate to my eyes. At the same time, it sounds like a well-read, undergraduate who doesn't fully understanding topics beyond the fundamentals. On the fundamentals, ChatGPT is surprisingly strong.
That's a summary of response based on my many hours of using it.
I suspect that there is something else going on than intelligence which will become obvious over the next few years.
There was a horse, "Clever Hans" who appeared to have the ability to answer surprisingly complicated mathematical questions. Did "Clever Hans" have mathematical intelligence. Not at all. He was responding to a cue unknowingly being given by his trainer.
I suspect the same thing is happening with ChatGPT. What if all that is happening is that the text is being formulated to very complicated cues that are implicit in the very complicated, statistical analysis?
A month or so ago I was doing some analysis on our mailing list traffic. I had a complex SQL query (involving tables mapping variations of email addresses to names, and then names to companies they worked for within specific date ranges), that I'd last modified a year previously (the last time I was doing the same sort of analysis), and didn't feel like wrapping my head around the SQL again; so I pasted it into GPT-4 and asked it, "Can you modify this query to group all individuals with less than 1% total contributions into a single 'Other' category?" The query it spat out worked out of the box.
Whatever it's doing, at least for code, it's not a glorified Markov chain -- there's some sort of a model in there.
I've always disagreed w/ Searle re the Chinese Room. My guess is that Searle never built an adder circuit from logic gates: combining irrational elements together into something rational is the core magic of computer science.
If you want to see someone asking humans questions where they consistently fail to be rational, to the extent that they sometimes seem to approximate a stochastic parrot, read Thinking Fast and Slow by Daniel Kahneman. (It might actually be interesting to give GPT-4 some of the questions in that book, to see how similar or different they are.)
I'm not sure why you disagree with the Chinese Room argument. I would be interested. I agree that Searle was solely a philosopher who did not take an engineering viewpoint.
Searle's main point is that if I have a book that tells me how to respond and I never learn Chinese, then I do not understand Chinese. If you see a flaw in this reasoning, I am very interested.
My point is just that LLM models are a compression of the content available on the internet equivalent to a rule book. It is definitely fascinating how powerful LLMs are as far as summarization and forming coherent responses to input.
I am a big fan of Kahneman and agree with you that it is will be very interesting to ask GPT-4 the questions in that book.
> Searle's main point is that if I have a book that tells me how to respond and I never learn Chinese, then I do not understand Chinese. If you see a flaw in this reasoning, I am very interested.
You don't understand Chinese, but you are not the process. For the process to understand, it doesn't require any single component to understand like some variation on the homunculus.
And while it might seem obvious that the bulk of understanding can't be contained in a book, you don't really have a book in the Chinese room. Not if the room does a competent job. You have some kind of information-dense artifact that encodes an enormous understanding of Chinese in an inert form. A sweeping library that covers uncountable nuances in depth.
Or to phrase it as a direct attack on the argument: The book does have semantics. You don't need qualia to have semantics, especially not the definition of qualia where nobody can prove they exist.
> Searle's main point is that if I have a book that tells me how to respond and I never learn Chinese, then I do not understand Chinese. If you see a flaw in this reasoning, I am very interested.
To a degree, I feel like the Chinese Room argument is begging the question. When I imagine Searle sitting in a room, with a book of instructions and paper and everything he needs to execute GPT-4's equivalent, I basically see an actual computer. That is literally what he is; there is no difference. So then to ask, "Does this system understand Chinese?" is literally exactly the same question as "Does GPT-4 understand Chinese?" You haven't actually illuminated the question in any meaningful way, except to give people not familiar with how microprocessors work a better intuitive understanding. (Which, upon reflection, probably is a fairly useful thing to do.)
I looked a bit at the "1990's version" of his argument on the Wikipedia page you quoted. Going back to my earlier example, this is sort of what his argument sounds like to me:
A1) Electronic gates just on and off switches.
A2) Numbers and addition are semantic.
A3) On and off switches are neither constitutive of, nor sufficient for, semantics.
Therefore, computers cannot add; they only simulate the ability to add.
Now I'm not up on the fine details of what "syntactic vs semantic" means in philosophy, so maybe #2 is't accurate. But in a sense it doesn't matter, because that communicates how I feel about Searle's argument: "I've made some distinction between two classes of things that you don't understand; I've defined one to be on one side, and the other to be on the other side; and therefore computers can't understand."
My best guess as to the "syntactic / semantic" thing is this: In some sense, even his premise, that "Progams are syntactic", isn't actually accurate: Computers operate on bits which are operated on by gates: gates and bits themselves don't inherently have symbols; the symbols are an abstraction on the bits. Even bits are abstractions on continuous voltages; and voltages are ultimately abstractions on quantum probabilities.
What a given set of voltages "means" -- whether they're numbers to be added, or words to be word-checked, or instructions to be executed, or a JPEG to be decompressed, depends entirely on how they're used. If you jump into the middle of a JPEG, your computer will happily try to execute it, and if you dump the program into your video buffer, you'll get a bunch of strange dots on your screen.
Furthermore, when you build an "adder" out of logic gates, you can build the gates such that they correspond to our intuitive idea of binary addition, with individual carries for each bit and so on. But this is inefficient, because then you have to wait for the carries to cascade all the way through the whole thing you're trying to add. Instead, you can brute-force a set of logic gates such that given these 16 bits in, and these 9 bits out (8 plus overflow), you just get the right answer; this will be a lot faster (in the sense that the signals have to go through fewer gates before stabilizing on the final answer), but the gates inside then don't make any sense -- they're almost a "compression" of the longer, carry-based method.
Does that mean that an adder made this way isn't "actually" adding? In the end it doesn't really matter: 16 bits come in, and 9 bits come out the way you want them to. It doesn't really matter that much what happened in the middle.
Putting all that together: It seems to me the "semantics" of a set of bits is based on how they end up interacting with the real world. If I can ask GPT-4 what's missing in my pancake recipe, and it can tell me "you're missing a leavening agent like baking powder", then it seems to me there must be semantic content in there somewhere, and all the arguments about syntax not being sufficient for semantic turn out to have been proven wrong by experiment.
Clever Hans is just a proxy. ChatGPT and other LLMs obviously can process information on their own. These two have nothing in common, even GPT-3 would have noticed this.
Let us disagree on what is "obvious". Given an input and an output, you believe that the complexity of the output proves that intelligence takes place.
I agree that ChatGPT is more than a proxy. Unlike Clever Hans, it is processing the content of the question asked. But it is like Clever Hans in that the query is processed by looking for a signal in the content of the data used to train ChatGPT.
The real question is where this intelligent behavior comes from? Why does statistical processing lead to these insights?
I believe that the processing is not intelligent primarily because I see that holes in the data available leads to holes in reasoning. The processing is only as good as the dynamics of the content that it being processed. This is the part that I believe will become obvious over time.
I thought you were saying it was "obvious" that the processing demonstrated intelligence.
My point was the level of intelligence shown is relative the quality and quantity of the data used for training. The data is where the intelligence is and the model is a compression of that latent intelligence.
The best summer intern that I ever hired did not have any programming experience. He was a Berkeley student who had recently decided to change his major but hadn't started taking any programming classes.
When we interviewed him, it was clear that he was a serious student, he was very smart, and that he would work very hard if we gave him the internship. As the hiring manager, I turned him down because of the lack of programming experience.
I was overruled by our CTO. Apparently, the candidate was a family friend of the CTO and the CTO had strong confidence that he would learn programming very quickly.
The Berkeley student was given the paid internship. In an 8 week period over summer, the candidate learned ios programming, identified a problem that was impacting customers, proposed a fix, and was able to release the fix to customers. When he demonstrated the issue and his fix to the engineering team, it was clear that he had spent long hours working with the app.
My lesson from this is that the best candidate may not be the one who appears the best on paper. More important is a very smart candidate who is willing to work very hard.
Who cares. They found a star. Steph Curry, Kobe Bryant, JFK & RFK, countless more were megastars and they had unbelievable legs up in their profession. But at the end of the day they performed like crazy.
Nepotism is bad when it promotes idiots to positions they can’t handle. It is GOOD when it ensures talent gets cultivated.
That the CTO was family friends with the kid was terrific luck. Everybody won.
The important part is that started from zero and went through all the steps to production in 8 weeks.
And I assume it is not the only thing he did, presumably he did that in addition to what he was hired for. When he found that bug, he probably did a good enough impression so that the company let him work on it.
Most interns introduce more bugs than they fix, and are usually assigned internal tools and other noncritical code for that reason. A net 1 bug fix is already a good score.
> The important part is that started from zero and went through all the steps to production in 8 weeks.
Why is that important? That may be personally impressive, but it's unclear how the company benefits from the intern starting from zero.
> And I assume it is not the only thing he did, presumably he did that in addition to what he was hired for. When he found that bug, he probably did a good enough impression so that the company let him work on it.
That's a lot to assume with no evidence. None of this was stated by the OP.
> An internship is just a glorified interview process. They found a good future employee
Why are you assuming that? The OP said nothing about the intern become an employee, much less a good employee. You'd think that would be mentioned if it were the case.
It is the definition of an internship. They’ve de-risked this hire. Now they can do things like offer a more competitive offer or otherwise close the candidate with high confidence
Short-sighted companies with high turnover probably won't see the benefits, companies that actually give their employees careers definitely will. It is all about potential, what is 8 weeks of internship if you expect your employees to stay for a decade or more? It may not be the culture of Silicon Valley, but the Silicon Valley culture is more the exception than the norm (at least for decent companies).
No, nepotism is covering someone who's _failing_ at their duties, yet retaining them due to familial connections.
Otherwise, what are you really mitigating by refusing to hire your CTO's nephew? If he's a strong candidate, it's not pathological and calling it nepotism is pointless at best. Otherwise if CTO's nephew turns out to be a weak candidate and is let go then again, it's completely normal. If CTO's nephew is a bad hire, yet he's retained, then it's nepotism.
> supports that the CTO was a better judge of the candidate.
No, it doesn't, because the hiring manager's choice never got a chance to show what they could do. Besides, the fact that the nepotistic hire worked out could have been just dumb luck. After all, hiring is a crapshoot, especially hiring interns.
Regardless, this was clearly nepotism, and the question isn't whether the CTO could judge the candidate, the question is whether favoritism was shown toward a family friend, which is indisputably the case.
That is a factor in why nepotism is so endemic - people are much better judges of the character of people in their family. This is a clear-cut case of nepotism, although personally I don't see a problem here. Nepotism isn't a bad thing in small doses. This instance is a good example of why not.
For me it would be nepotism if CTO would hire a family friend that was totally unqualified or somewhat unqalified and would cover his ass to keep him in company.
Story in here is not nepotism, it is just that guy had more opportunities than others but that is just the way life is. That is why for example going to university is important - not for knowledge or lectures - but for getting to know people who will most likely be working in the same field.
No, nepotism is covering someone who's _failing_ at their duties, yet retaining them due to familial connections.
Otherwise, what are you really mitigating by refusing to hire your CTO's nephew? If he's a strong candidate, it's not pathological and calling it nepotism is pointless at best. Otherwise if CTO's nephew turns out to be a weak candidate and is let go then again, it's completely normal. If CTO's nephew is a bad hire, yet he's retained, then it's nepotism.
(is kin) -> (insider knowledge that they're better than expected) -> (preferred for job)
Specifically,
(insider knowledge that they're better than expected) -> (preferred for job)
would make complete sense and would not be looked down upon. The source of that knowledge is problematic on a societal scale, but not on an individual level.
Person A and B are equally great for the role but cannot do a triple summersault to land on a beam, so neither can get past the interview stage
But… A’s uncle who went to Harvard with a high-up says “that boy is good” so they hire him on that. B posts on HN about getting no feedback and submitting their CV to 100 firms.
Nepotism has happened.
Later A may/may not succeed, turns out he does in this N=1 case.
This debate about definition is sort of irrelevant to the broader question. Should knowledge that someone is good be ignored if the knowledge was gained through prior personal knowledge of a candidate? Even if that was possible, it would seem to me an absurd a route to take. Surely the better question is how those without prior connections can be allowed to sufficiently demonstrate their competence.
This is what nepotism is. It does not have to be that the person benefiting from it is incapable. Most of them are normal capable, some better then average some worst.
> It's like saying that grocery store's owner's daughter shouldn't be hired at their grocery store.
We're all here trying to tell you that the grocery store's owner hiring his own daughter is literally, indisputably, paradigmatically nepotism, but somehow you don't seem to get it.
I don't think anyone is arguing the owner (as an individual) is immoral. Denying the daughter that won't change anything, and teaching his daughter life skills is an even higher responsibility.
Rather, we are lamenting the inherent unfairness that is often overlooked or not acknowledged.
> In an 8 week period over summer, the candidate learned ios programming, identified a problem that was impacting customers, proposed a fix, and was able to release the fix to customers. When he demonstrated the issue and his fix to the engineering team, it was clear that he had spent long hours working with the app.
This is actually faint praise that this was the right decision. Finding and fixing a bug (there is no mention of how tricky or complicated it was) is often the warm up project for an internship to get them familiar with the code base before doing something more ambitious.
The fact that the presentation demonstrated “spent long hours working with the app” and not some particular insight or ability is also telling.
As others have mentioned this whole story smells of nepotism.
How does one filter for candidates who are very smart and willing to work very hard without having a trusted person already at the company to vouch for them?
> How does one filter for candidates who are very smart and willing to work very hard without having a trusted person already at the company to vouch for them?
Experience? ;-)
"As the hiring manager, I turned him down because of the lack of programming experience."
This candidate was already filtered out, but the CTO put him back in.
Let me put it this way: if companies can't figure out who to hire, then the fairest selection method wouldn't be nepotism, it would be random. Give everyone an equal chance to prove themselves.
>if companies can't figure out who to hire, then the fairest selection method wouldn't be nepotism, it would be random
Companies don't optimize for fairness, they try to maximize quality on their end. And many, including myself, would argue that nepotism as flawed as it is, is still superior to random selection.
I am baffled when smart people say something like this. Without consciousness, without a stake in the game, the behavior is pure statistics. Statistics is limited by probabilities and logical gates. Nothing else. Consciousness is about being aware which means insights about context and harm. People are limited in ways that a statistical engine is not and that makes all the difference in the world.
> What I am trying to say is, it doesn't matter if the subject is actually conscious. All that matter is we (I), feel/think that it is conscious, or deserving of care and respect.
My point was that it does matter if the subject is actually conscious. Human beings are easily fooled and it does matter if people mistakenly think it is conscious when it is not.
You are possibly mistaking free will with being an agent that has a (possibly deterministic) method of updating priors in response to new (unknown to the agent, and possibly deterministic) input.
I apologize for not being more clear. I find it very challenging to distinguish between details about LLP and "consciousness". My key point is that "human consciousness" is very different than ChatGPT. ChatGPT is statistically processing content already created and "appearing" to be conscious. Human beings have characteristics that ChatGPT does not share (sentience and context). We are often mistaking the what is needed to generate content (human consciousness) with what is capable of processing that content in very interesting ways (ChatGPT).
I do not believe that I am confusing free will and consciousness. See my comment above. Determinism versus free will is independent of knowledge available. Consider a paralyzed person incapable of any action. That person if the senses are all working still has awareness and context. A statistical engine only appears to. A LLM model is basing all actions on a complex matrices of thresholds. It is surprising and amazing how well that works. Given stimulus that takes advantage of those minute differences in thresholds, a wrong response will be returned. Human are not fooled in this way. Minute differences are typically missed or even skipped. Human beings can be fooled by optical illusions and by contradicting context (a statement like pick the "red" circle written in green ink and the person mistakenly picks the "green" circle. LLM models do not make these types of mistakes.
Are you sure? That's it? I seriously doubt it. Your claim too heavily relies on a presupposition of what is to be shown. Somewhere between the four fundamental forces and us is zillions of light years of unexplored blue sky. Please meet the rest of us there
Not really clear what you are unclear about. My main point is that human beings have sentience and can reason about cobtext in terms of how an action affects others. Computers are following a statistical algorithm without any sentience or any understanding beyond the statistical thresholds. Ignoring the complexity and brilliant mathematics, it is at the core no different than a key word matcher like the classic application "eliza". Its performance is amazing but it is really the same algorithm at its core.
Murdoch was successful in many markets before starting Fox in the US. His formula was to go after the tabloid market with higher quality content (plenty of stories on scandals and beautiful/famous people, some conspiracy theories, and also quality content). Conservative media had been dominating talk radio and there was plenty of conservative media (Heritage Foundation, National Review) for conservative viewpoints before Fox. CNN (via Ted Turner) had shown that cable news could be profitable and Fox/Talk Radio have shown that rage media creates a loyal audience.
It's not that mainstream media is liberal media as often claimed (I find much of establishment media: CNN, MSNBC, USA Today, People, etc. too superficial to be liberal or conservative -- it's just a business that tries to attract the establishment audience which tends to skew liberal). The real issue in my mind is that mainstream media has not been very good at rage media while Fox and Talk Radio have been virtuoso at it. Trump temporarily made it easy for CNN and MSNBC to thrive at rage media (all liberals could agree on their shock about Trump's actions), but without Trump, liberals naturally fall into infighting between progressives and establishment viewpoints (for example, Bernie Sanders versus Hillary Clinton or AOC versus James Carville)
The left has definitely learned rage media. Even after Trump. Musk is now the baddie of the month. Any other wealthy white male can easily fill the void when we reach peak Elon Musk clickbait. Nearly all world events are explained by racism or sexism. The entire woke movement is a direct result of left leaning news orgs battle for profits
I am not saying that the left does not get outraged. I am saying that the mainstream media has not been able to capitalize on that rage for profit. That seems to me to be more the issue than an over-saturation of the liberal perspective or a lack of over-saturation of the conservative perspective.
Talk radio and Fox news are highly popular. As I understand it, before he was fired, Tucker Carlson was the most watched news-related show.
I agree that there are popular progressive and left-leaning shows that focus on outrage (John Oliver, Jon Stewart, etc.) but as I understand it, they are not as successful or popular as Fox News and conservative talk radio.
My point was not that conservatives are more outraged than liberals (I suspect that each side sincerely believes that the other sides shows greater outrage). I just meant from a business perspective, it appears to me that the establishment cable news services are struggling to maintain audience loyalty which tells me that they are not as good as conservative media at taking advantage of audience outrage.
I keep hearing this about abstract mathematics "never" having an impact because it is too abstract and relates to pure mathematics. It's not true. Mathematics is a formal system that provides insights on surprising patterns. Surprising patterns can almost always be applied outside their intended area. And not surprisingly, non-Euclidean geometry and even the inability for mathematics to find certain proofs related to primes has resulted in breakthroughs in other areas. Surprising patterns take time to have effect mostly because they are not generally known until some genius is able to apply them outside their intended area.
I am 99.999% certain that you are wrong to say that "this kind of abstract mathematics will 'never' have any meaningful impact on day-to-day software engineering." I would be less certain if you replaced "never" with "will probably not have a significant impact in the short term".
I understand that there are sometimes deep connections between seemingly unrelated fields of mathematics. However, are there examples of applications of really pure abstract mathematics in the fields of say, symplectic geometry or differential topology (or any other very abstract fields), that have come to impact the lives of a significant number of software engineers?
Most programmers (myself included) spend their days using IDEs building a new API in a CRUD app or implementing a new UI widget in a mobile app. I may be unimaginative but I have a hard time seeing modular forms, fiber bundles, or exact sequences making a breakthrough in my life and impacting my programming.
Agreed. That's like hearing about early group theory and disregarding it as abstract nonsense. Turns out it's pretty useful everywhere now, but at the beginning it was not so obvious.
I would also change the statement to "never have a direct impact on programming". Indirectly in 100 years this may lead us to a better understanding of physics (modular forms have some weird hypothetised connections to physics that I don't really understand), which may let us construct better computers, which will obviously have an impact on programming.
* Mindset by Carol S. Dweck
* Innovating: A Doer's Manifesto for Starting from a Hunch, Prototyping Problems, Scaling Up, and Learning to Be Productively Wrong by Luis Perez-Breva
* Fall In Love with the Problem, Not the Solution: A Handbook for Entrepeneurs by Uri Levine
These three books really changed my viewpoint and I've been rereading them every year.