But in a world of subscription based services, it has to ship more than once. And as soon as that's a requirement, all of the above applies, and the LLM model breaks down.
Is it still getting better? My understanding is that we're already training them on all of the publicly available code in existence, and we're running in to scaling walls with bigger models.
I pull messy data from a remote source (think OCR-ed invoices for example), and need to clean it up. Every day I get around 1k new rows. The way in which it's messed up changes frequently, and while I don't care about it being 100% correct, any piece of code (relying on rules, regex, heuristics and other such stuff) would break in a couple of weeks. This means I need at least a part time developer on my team, costing me a couple of thousands per month.
Or I can pass each row through an LLM and get structured clean output out for a couple of dollars per month. Sure, it doesn't work 100%, but I don't need that, and neither could the human-written code do it.
Effectively, LLM resulted in one less develper hired on our team.
It resulted in one less developer, but you're still using that tool, right? Didn't a human (you) point the LLM at this problem and think this through?
That fired developer now has the toolset to become a CEO much, much easier than pre-LLM era. You didn't really make him obsolete. You made him redundant. I'm not saying he's gonna become a CEO, but trudging through programming problems is much easier for him as a whole.
Redundancies happen all the time and they don't end career types. Companies get bought, traded, and merged. Whenever this happens the redundant folk get the axe. They follow on and get re-recruited into another comfy tech job. That's it really.
Human or LLM the trick with messy inputs from scanned sources is having robust sanity combs that look for obvious fubar's and a means by which end data users can review the asserted values and the original raw image sources (and flag for review | alteration).
At least in my past experience with volumes of transcribed data for applications that are picky about accuracy.
I do! I posted this a while down below but I guess the way the algorithm here works it got super deprioritized. Full repost:
I can chip in from my tech consulting job where we ship a few GenAI projects to several AWS clients via Amazon Bedrock. I'm senior level but most people here are pretty much insulated.
I think whoever commented once here about more complex problems being tackled, (and the nature of these problems becoming broader) is right on the money. Newer patterns around LLM-based applications are emerging and having seen them first hand, they seem like a slightly different paradigm shift in programming. But they are still, at heart, programming questions.
A practical example: company sees GenAI chatbot, wants one of their own, based on their in-house knowledge base.
Right then and there there is a whole slew of new business needs with necessary human input to make it work that ensues.
- Is training your own LLM needed? See a Data Engineer/Data engineering team.
- If going with a ready-made solution, which LLM to use instead? Engineer. Any level.
- Infrastructure around the LLM of choice. Get DevOps folk in here. Cost assessment is real and LLMs are pricey. You have to be on top of your game to estimate stuff here.
- Guard rails, output validation. Engineers.
- Hooking up to whatever app front-end the company has. Engineers come to the rescue again.
All these have valid needs for engineers, architects/staff/senior what have you — programmers. At the end of the day, these problems devolve into the same ol' https://programming-motherfucker.com
Yes, incredibly small scale data. I can find someone somewhere who has experienced just about anything. Hell, they don't even have to experience it - they just need to say they do.
I've known multiple people who swore potatoes cured their sickness. Yes, potatoes. No, not eaten - worn.
My sample size is greater than yours. Does that mean potatoes cure illness purely by being in proximity to people? Probably not.
People are dumb and unreliable. People are fueled by beliefs. And when those beliefs are challenged, almost every time this is taken as PROOF of the belief. They are untouchable. Agree with them and that's great, disagree and you're being silenced or the government or something thereby giving validity to their belief.
Sure. Now dismantle “scientific” papers published by government agencies with a political motive.
This is beside the point however. Both the government studies and the boots on the ground in fact agree that masking causes significant developmental disabilities, the only disagreement is the extent.
I don't necessarily disagree. But I am extremely wary of people who have a "government bad" mentality. It's very easy for such people to slip into outright science denial. And then suddenly the Earth is flat, covid was planned, climate change isn't real, and 5G controls your mind.
Yes, things can have a political motive. But you ALSO have political motives, you just don't know it. For example the climate change denial has a very real political motive - to protect the oil and gas industry. Oil and gas have been denying climate since the 60s, and our government backed them because of political will.
Masking, during the pandemic, saved lives. This is not up for debate. Whether it's a good idea now is another question all together.
I don’t really have much interest in engaging with your “you think X? you probably also think X Y and Z” strawmanning. It’s bad logic and makes for poor debate.
As for masking during the pandemic, it might have save some lives. It probably did. But it also contributed to long term damage of many others. Refraining from sniping an active gunman “saves lives” too, but that’s hardly the end of the discussion. Context matters.
That's not what I'm saying. What I'm saying is that simply distrusting the government isn't enough. It isn't enough to support any stance. You need more than that.
> As for masking during the pandemic, it might have save some lives. It probably did. But it also contributed to long term damage of many others
I mean... no, this just isn't true. Masks didn't hurt anyone because they literally can't.
This is what I'm talking about. The reason people didn't want to wear masks isn't because they thought they were harmful, but rather because they were TOLD to. They didn't have anything past that.
Even in the worst-case scenario where masks do nothing, there's still zero reason not to wear a mask. the only reason is "gov bad". That, to me, is not enough. That's a nothing-burger. But millions ran with that. There was also some misinformation about CO2 buildup or something but, go figure, that was entirely made up by alex jones types to justify the fact they didn't want to wear a mask.
Have there been any actual documented instances of this? I see it cited a lot, but rarely any specific names of DAs or cases attached to it, and it feels like an easy excuse for police to not even attempt to do their jobs.
Even without a specific policy to not prosecute crimes, there's a pipeline problem.
Courts are backed up and jails are full. DAs have to prioritize cases so that important cases are prosecuted within the statutes of limitations and constitutional speedy trial limits. Arresting people under suspicion of crimes that are unlikely to be prosecuted isn't very effective and nobody likes doing ineffective work.
or everything goes perfectly but the punishment is time served, a fine they can't pay anyway or some sort of service & promise not to do it anymore. I believe police and prosecuters want to do good work, but they also want the biggest impact of their work.
You can disagree with me all you want, but this guy was recalled as SF DA for letting criminals go.
His wikipedia enumerates several cases of him doing just that.
> In March 2020, Boudin charged 20-year-old Dwayne Grayson with elder abuse after he filmed 56-year-old Jonathan Amerson in February 2020 swinging a metal bar at an elderly Asian man in Bayview–Hunters Point, San Francisco and stealing his aluminum cans. Amerson was charged with elder abuse and robbery. The video later went viral online. Boudin dropped charges against Dwayne Grayson after a spokesperson in Boudin's office said the victim expressed his intent to pursue restorative justice.
> Boudin has been criticized for his alleged lack of prosecution of drug-related crimes, with only three drug convictions in 2021, none of which were for fentanyl dealing. Boudin has defended his actions saying that many of the drug dealers in the Bay Area are from Honduras, and would face deportation if convicted of drug dealing.
Except that even when the DA was trying to pursue stuff, the SFPD refused to help, and forced the DA's office to do stuff like move evidence in a U-haul truck.
I agree with you on this one case. But this was after years of him letting criminals go. This case was in the run up to the recall in which he was removed from office by SF voters for being too soft on crime.
That was the narrative that was presented. However, though the data made available by the DA's office is limited and flawed, I don't think the story is really borne out.
After a little more than a year in office, she had raised the conviction rate slightly, but had a lower charge-filling rate -- i.e. charged fewer things but won more.
> in her first 15 months, raised the city’s conviction rate for the first time in eight years, according to data from her office.
> Despite the reversal in conviction rates, it appears that Jenkins is taking fewer individual cases to court. Her charge-filing rate is about eight percent lower this year than last.
> While conviction rates have risen slightly, the total percentage of cases charged, or prosecuted, by the DA has remained relatively flat.
This also came with a shift towards more convictions for petty theft and narcotics, and less use of diversion programs. But given that the charging-filing rate actually decreased, I think this implies that other more serious stuff was being charged less. So the "soft on crime" guy actually was filing charges at a higher overall rate, and was prioritizing the more serious crimes.
Note, I tried to view the DA data dashboards today and none of them even load for me. The article above shows top-line rates but not with a breakdown by the type of violation.
"Please don't comment about the voting on comments. It never does any good, and it makes boring reading" [1].
> guy was recalled as SF DA for letting criminals go
Your comment would be stronger if you cited something from the article, versus just presenting it unadorned. (And at risk of being a hypocrite, I upvoted your comment.)
In the US, there's no evidence that this happens - just a bunch of media narratives and failed lawsuits after the 2020 election that couldn't provide an ounce of proof when push came to shove.
Eh, there's voter fraud in the US. Some of it is very ironic too [1].
> He voted by absentee ballot and again in person on election day but claimed in social media posts that he did it to show how insecure absentee voting is. He pleaded guilty to one of the charges and was sentenced to 6 months of probation and ordered to complete 40 hours of community service and pay $500 in restitution. [1] [2]
However, I didn't see any cases for 10+ votes as a deceased person for 2020. There is somebody that voted 26 times using alive people though [3].
The overall moral is though, there is fraud and it does get caught. When you think there's "widespread" fraud that isn't being detected it ends up always being a simple explanation (i.e. people lived in the state at the time of the election and moved afterwards).
I've always seen this as the exact opposite view - from go's concurrency model, every function is "synchronous" so the caller is not given a choice, if they want to run it asynchronously they have to create a new thread, then if they care about the result deal with inter-thread communication.
With async/await, you're explicitly giving control to the caller to decide, you can await this promise now and have the thread treat it as synchronous, you can spawn a new task to run it in the background, or you can join it with other promises where you don't care about order and await for the results of the group.
1) Mental model. My claim comes from the firm belief that the more code is aligned with how we think, the easier it is to reason about the code. I naturally think about actions as they are not async or sync by nature – rather, it's me who's in charge of how the action is going to be executed (back to my "watching TV" example). Human attention here serves as an analogy to utilizing the logical CPU core during runtime.
2) Performance consideration. What you described indeed can work, too, but it comes at a cost. With Go, yes, you have to handle async results yourself (if you care about results), but you now understand the price of this and can make better judgments of the code and complexity and have better performance overall.
Isn't the entire point of a public health crisis like this that you can't just focus on yourself and your family, because doing so still has the potential to seriously impact or kill other people?
> Isn't the entire point of a public health crisis
Ha, I like the wording. As if there's a point to a public health crisis. I mean, yeah this is _the_ point if the powers that be are orchestrating this.
