This is neat -- how would you think about evaluating the quality loss as you change to more efficient models? I saw you did an analysis on the number of messages, but wondering if there's more robust methods?
In offline training of our router, we run extensive cross-domain evaluations to determine when a smaller model can handle a request without any quality loss relative to more powerful models. In an online setting like our chat app, there's probably some more rigorous post-hoc analysis we could do on response quality—could make for a good follow-up post.
Iranians are willing to die (and are dying) to fight the current government. Inflation and the economy has destroyed their hope of a future for decades. Talk to any Iranian now and they’re all hoping for the end of the regime. It can’t get much worse.
Iran is not Syria, there’s a flourishing academic population, the fundamentalists are the ones in power, plenty of people still remember what life was like pre-1979.
Regime rapes and murders women and children, imprisons thousands, mismanages the economy to the point where millions live in poverty and everyone's life savings are slashed in half , environmental degradation to the point where lakes are drying up, state-sponsored terrorism etc....but hey guys have you "pondered" how bad state collapse would be?
Method matters. If people overthrow the government in violence, then it is just another turn of the revolving door of violence. What will happen to the former partisans and supporters of the old guard? Not to mention, if a movement without sufficient ability to take power rises and fails, the loss in life and chaos in society will be far worse than the status quo.
Iran is not starvation level desperate, nor people trapped under earthquake rubble desperate, to my knowledge, so I do not expect a successful popular movement without buy in from the elite and military classes of the country.
Yeah, I’m sure if we would have just waited a few more years, slavery would have also ended, we would have ended child labor, women would be allowed to vote… come on, this is just utterly dumbly naive.
Can't say much about slavery, but the issues of child labor and women rights weren't solved through a bloody revolution, and there are good arguments that they were primarily the consequence of technological progress fueled by economic growth.
On that note, we don't know for how many decades a popular uprising will be even possible. Technology may offer a turning point at sometime, where a rogue government can obtain total control and force over its populace.
Actually Patience might be the best answer. The folks leading Iran are all quite old and once the big guy dies - only a few more years - change will happen naturally.
Don't need millions of deaths in another revolution.
Please assume good faith and don't incite flamewars.
When I'm arguing against wishing for collapse "from the comfort of our countries", where do you read that I'm talking to Iranians? I'm talking to HN commenters who are not from Iran. I'm obviously not arguing with pinkbeanz since he/she is from Iran.
Try arguing in good faith, unless you want this to devolve into a flamewar. You already tried something like this in another comment: please stop.
> You're in the wrong here, it's time to walk away.
Nope. The people who are in the wrong are the coach warriors who want to see collapse but aren't unwilling to go fight for it, making arguments from the safety of their countries.
How so? I'm not calling for any specific course of action, I'm just saying "be thoughtful before arguing in favor of collapse", especially if you don't live in the country under discussion.
My "weapon of choice" is calling for restraint before violence.
That's the go-to lie of people trying to start civil wars and scare people into taking their side. "If you're not with us, you're against us". "Not taking side is just supporting status quo, which is taking a side". War is peace.
It reads to me like others are talking regime collapse, whereas you're concerned about societal-level collapse. I'd agree few with good intentions would wish for the latter.
Regimes can and do collapse without it meaning anything like the level of nationwide chaos and misery you seem to have in mind.
Yes, I think that's what may be happening, at least for the people arguing with me in good faith.
In that case, let me be explicit that I'm pessimistic about the current government of Iran (a dictatorship which I'm not in any way in favor of) collapsing without a lot of destruction and bloodshed, and that I do not wish this destruction on the families currently living there. I also fear that if this collapse leaves a power vacuum, something like -- similar, not exactly the same -- ISIS could rise in its place. I'm very pessimistic about this, and so I'm wary of wishing for collapse.
I interpreted GP as talking specifically about induced regime collapse, specifically one induced externally. As in, not about the regime itself slowly decaying into a more benign form, but rather about attempts to remove it by force.
Historically, I can't think of a single case of a regime being destroyed through revolution or invasion that didn't end in at least partial societal collapse and a drastic increase in deaths and suffering for a generation or more.
The Nazi Reich? Arguably it brought about its own downfall via its misguided attempts at "invasion" - at any rate it was about as violent and sudden an end to a regime as you could ask for. But history is full of coups d'etat that didn't necessarily negatively impact the greater population all that directly.
You're right, Nazi Reich is a valid counterexample to my assertion. It was destroyed by external forces, though as you say it was totally self-inflicted, and a lot of people living under it suffered greatly in the process, but the situation for them improved very quickly.
Per my understanding of history, that last part was an anomaly in several ways. This being a world war is one way, of course, but another factor was that the hot war between the Allies and the Axis transformed, after the Axis was defeated, into a cold war between members of the Allies. Both sides of this new conflict considered it critical to capture and stabilize the very territories they helped liberate from the Nazis. This wasn't a half-hearted "nation-building" program like we've been seeing in more recent times - both the US and USSR committed tremendous amounts of resources to get Europe back into shape, because this was still a war - arguably the same war, just going through a cool-down period - and both sides expected it to go hot eventually.
Also, while Germany survived the death of Nazi regime quite well, the Cold War is also known for US and USSR sponsoring and orchestrating coups and regime changes all around the world, and (AFAIK) those cases all ended badly for the locals.
In some sense, it might be that World War II was itself an anomaly - I can't think of any other war that ended with both the winners and the losers coming out better off. But it's also worth remembering that WWII itself was in large part a consequence of the societal collapse Germany underwent after losing WWI. And the subsequent Cold War was in large part a consequence of societal collapse caused by bloody revolutions in Russia and elsewhere around the start of the 20th century.
The way I see it, we have one special case of regime collapse making everyone better off almost immediately, but even that one is surrounded and infused with countless cases of regime changes that caused generations to suffer.
I asked ChatGPT for some other examples - it listed "the overthrow of the Somoza dictatorship in Nicaragua in 1979 [by] the Sandinista National Liberation Front" and "the overthrow of the Marcos dictatorship in the Philippines in 1986".
With additional prompting it also listed the overthrow of the apartheid regime in South Africa, and the Solidarity movement's ultimate success in Poland in 1989.
"good faith" - it is a highly subjective term, but it is typically used as if it is objective.
There is a whole class of reality distorting phrases like this in Western culture[1], this sort of thing has always been with us but seems to have taken on much more causal significance with the rise of the internet.
Instead of replying indirectly, please address what I'm saying: what's the actual "reality distortion meme" I'm deploying here? Be upfront and accuse me of something I can defend myself of.
> "good faith" - it is a highly subjective term
HN defines is pretty clearly (note there's more, I'm just quoting some parts):
> "Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes."
So snark replies are out.
> "When disagreeing, please reply to the argument instead of calling names."
So calling someone an Iranian secret police agent is out.
> "Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith."
So instead of cross-examining me or trying to "catch me" somehow, address the fact I'm calling for nonviolence and restraint, and that I claim recent experience in the Middle East shows that regional collapse leads to the rise of fundamentalist groups and a general rise of unchecked violence. Assume good faith; assume I want the common good. If I made a mistake, reason with me. If you are an Iranian, don't withhold this information from me until we are 10-levels into a nested discussion.
> "Eschew flamebait. Avoid generic tangents. Omit internet tropes."
Self-explanatory. I'd say name-dropping "reality distortion memes" is one such internet trope (one, to be frank, I still don't understand because you haven't explained).
> Instead of replying indirectly, please address what I'm saying: what's the actual "reality distortion meme" I'm deploying here? Be upfront and accuse me of something I can defend myself of.
"good faith" - it is a highly subjective term, but it is typically used as if it is objective.
I will copy/paste this every time you represent that I have not disclosed the term - to others that sort of thing might be annoying, but to me it is fun!
>> "good faith" - it is a highly subjective term
> HN defines is pretty clearly (note there's more, I'm just quoting some parts):
>>>>> "Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes."
> So snark replies are out.
a) People break the guidelines all the time.
b) "Be kind. Don't be snarky...." - this text stands on its own in the guidelines and is not given as a definition of good faith.
c) The only reference to "good faith" in the guidelines is this (which you are in violation of, as am I (and I have strong ideological reasons for my non-compliance)): "Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith."
d) There is an important distinction between the definition of a term, and each individual's classification of behavior as being a valid instance of the term.
> So calling someone an Iranian secret police agent is out.
Agreed, thus I have not done that.
> So instead of cross-examining me or trying to "catch me" somehow, address the fact I'm calling for nonviolence and restraint, and that I claim recent experience in the Middle East shows that regional collapse leads to the rise of fundamentalist groups and a general rise of unchecked violence.
I acknowledge that you believe this, and that there is surely some truth to it.
I will not refrain from criticizing your claims though.
> Assume good faith; assume I want the common good. If I made a mistake, reason with me.
I will assume what I want to, or nothing at all.
My reasoning is above.
> If you are an Iranian, don't withhold this information from me until we are 10-levels into a nested discussion.
I am not Iranian.
> Self-explanatory. I'd say name-dropping "reality distortion memes" is one such internet trope (one, to be frank, I still don't understand because you haven't explained).
"Good/bad faith" is also a (much more) popular trope, one that I believe is also much more dangerous.
This reads like a low-key, state-sponsored bot post on Twitter: spreading FUD under the guise of being “safe.” Iranians need to take back their country, they are already not safe.
At this point, the only transfer of power that will happen will be due to a hostile civilian takeover. The current regime will not abdicate and any foreign military coup will likely not be tolerated by the people.
A revolution will serve as a cautionary tale to future governments who attempt to push fundamentalism as law.
One is right to be careful about another state collapsing, after the failures of Irak, Syria, Lybia, etc....
However, I also think the Iranian situation (and unlike many commenters I know Iran, have been there many times, and more) is different because there already is a quite successful diaspora of Iranians ready to take back control of the country.
Iranians have managed to somehow maintain a decent amount of infrastructure despite the sanctions and the general adversity (they did not become Cuba). There is an inherent vitality within the Iranian people which makes me very confident that once the islamists are rooted out, their country will spring back to being a working economy thanks to its intellectually very capable people and its diaspora.
I hope what you're saying actually results in a transition with minimal violence. I would also like minimum interference from foreign powers, the US included.
I think you’re confusing giving the company a chance to respond with made-up news. The time a news outlet gives for a response is usually proportional to the severity of what they’re announcing.
Giving a company a chance to respond to some hiring numbers when they likely have multiple anonymous sources who have been credible before is a different bar than asking for comment on something that alleges fraud or illegal activity.
If a company doesn't respond to the made up fantasy, that doesn't suddenly make the made up fantasy any more real or proven.
> The time a news outlet gives for a response is usually proportional to the severity of what they’re announcing.
This is an assumption with 0 substantiation. Actually in 2023, I would say the average main stream media outlet does NOT get the benefit of the doubt here.
> likely have multiple anonymous sources who have been credible before
This is an assumption with 0 substantiation. Again.
Why does this media outlet deserve the benefit of the doubt? I just don't get it.
It’s common journalistic practice. That’s one of the benefits of reporting in mainstream media, they have pretty high standards for what gets published.
That doesn’t mean they’re always right or there aren’t bad journalists, but if you spend time listening to most journalists explain their reporting they often explain exactly how they verify their stories, what they went through to get sources, and what they were and weren’t able to report.
Journalism is one of the few ways we have in a democracy to hold power to account. What I don’t get is what a lack of faith in this institution does for us except give those in power even more ways to be unaccountable for what they do.
Forgetting the home vs fast-charger nuance here, important to remember than electric vehicles have no spark plugs, no muffler, no engine oil, and a very simple coolant system.
My Volvo XC40 Recharge has a 2-year/20,000 mile maintenance cycle, the only work it needs is a check of the coolant and brake fluids and replacing cabin air filters.
Having suffered through broken engine mounts, torn exhaust flex pipes, painful spark plug replacements, coolant leaks, pump failures, and countless other issues with ICE engines, the simplicity of electric is a breath of fresh air.
Of course there’s the question of what happens to the battery over the long term but i’ll take that trade over what ai had before any day.
I tend to agree, though on the flip side there's things like the octovalve, which is a construction that doesn't really exist on an ICEV. I think there are some trade-offs, though I expect that over the long haul we'll see that EVs are generally more reliable. For certain I think that the norm will be the battery lasting the life of the vehicle, and when it does get replaced it'll be about as bad as replacing the drivetrain in an ICEV. So a wash in that regard.
Any ICE car only has in addition 2-3 oil changes in that 20,000 miles maintenance cycle. Spark plugs are 100k miles.
Your car doesn't have an engine mount, but it has a motor mount that could fail. You have already admitted to coolant, which implies it could leak and it has a pump that could fail.
Most people who complain about how bad EVs are seem to be comparing a new EV to a very old ICE, or memories of their parent's generation which wasn't as good as modern ones. New ICEs are reliable and have been for a few decades.
I’m really comparing it to my previous ICE cars, both an Audi sedans and a Mercedes. All built in the last 10-15 years.
100,000 miles for spark plugs is extremely rare. I just pulled up a 2022 A4 and it’s 30k. Transmission fluid is 40k miles. I don’t think it’s a stretch to say that the heat, vibration and weight from an ICE engine wears down parts faster. Add in complexity from turbochargers and there’s even more parts that can fail.
The coolant system on an electric vehicle is extremely simple. There’s no radiator even. An ICE coolant system is far bigger and more complex.
That’s right, they’re solving the same problem, although metric/semantic layers have traditionally lived within BI tools. Looker’s LookML is a great example of a metric layer.
There you define the relationships between tables, which dimensions to expose to your users, and how measures are to be aggregated along which dimensions and at what roll ups. Filters, formatting, and other semantic information rounds out what’s possible, making it easier to build complex reporting without having to pre-build large joined tables in the warehouse.
In theory a semantic layer decoupled from BI can be useful too, as downstream systems can all query a metric. Maybe an operational workflow needs access to revenue numbers or a spreadsheet wants a feed of some fresh aggregates.
The real question is will BI tools adopt a semantic layer they don’t own, giving up control over a key part of their input into their visualizations? Time will tell.
Oracle BI has this separation in the RDP layer. So there is some history of BI tools separating out that layer but the separation doesn't have to result in giving up control. It's more of an organizational decision. Although it's probably not advisable to have wide open access to the semantic layer to the same users creating dashboards, as things can quickly get overcomplicated and hard to maintain.
I'm curious what this enables can't be done with a thoughtfully-designed data warehouse. Looking at the "With MetricFlow" and "Without MetricFlow" examples, it seems like it's designed for people who are willing to use a new query language as long as it's not SQL?
On the one hand, the compile targets for MetricFlow and DBT are SQL. So strictly speaking, nothing can be done with these that can't be done with a data warehouse.
SQL doesn't really have a way to define new aggregations well. So taking something like Daily Active Users / Monthly Active Users (DAU/MAU), which is a semi complicated calculation using windowed data, there's no good way to have a repeatable DAU/MAU aggregation concept you could reuse at different levels without having to re-write that sql that defines the aggregation. EG DAU/MAU against the account, or DAU/MAU by user type, or DAU/MAU by region would all be different queries/reports.
Instead if you have a layer that can be given the definitions of how a particular aggregation is structured, and it knows how data is related to each other, you can ask for that metric against different concepts or levels without having to repeat yourself.
Extending that out, if BI tools can understand that layer, then you can get consistent WYSIWYG report building across multiple tools without having to redefine the calculations everywhere.
In practice - complicated time series metrics, especially on top of derived temporal-logic attributes like funnels, activation, etc., are phenomenally difficult to write by hand in SQL for most analysts. We are in the process of switching to dbt metrics and it cuts the effort down for this kind of thing 5x-10x, and the SQL code dbt generates runs signficantly faster too.
Looking at the DBT semantic layer documentation I think I get it. The goal seems to be describing a metric that could be aggregated at multiple levels once instead of defining all of the aggregations separately.
For example, if you want to be able to aggregate revenue by subscription plan and by country over year, month or day, then you would have 6+ different aggregations. The more dimensions and time periods you add, the more the various combinations blow out. (I think the formula is dimension permutations multiplied by the number of time periods.) Having a single definition of the metrics prevents errors from occurring in rote implementations of the metrics. A bit like generics, but for queries.
At this point, I'm curious how they manage to make the integrations with different platforms work. (Ultimately, I think that's why they acquired Transform.)
I think that’s still the outstanding question. Tools like Mode and Hex have already started supporting dbt’s metrics layer, but will big players like Looker and Tableau also move to support it? That’s harder to say
One of Transform's strong suits is their Tableau integration - they have really good tooling for pushing metrics out to Tableau, rather than relying on Tableau to pull them in. Apparently AirBnb was/is still a very dedicated Tableau shop.
Minerva actually had really poor tableau support as of a year ago, that is something Transform improved on. Most of Minerva consumption was done via Superset, GoogleSheets, and email
Believe the replacement was because bash 4 was GPL3 licensed, and zsh was not. It wasn’t so much for user benefits than it was for security and licensing.
