I strongly doubt that the campus is any more of a force multiplier than a collection of far cheaper class A office buildings of similar capacity, with similarly serene grass and vegetation connecting them.
768 karma · joined January 20, 2012
If you think in long timeframes, want to build for the future of software, join us at kinelo.com
Generally happy to chat (except recruiters...): keith [at] kinelo.com
Even within a single project of a single team, the remote vs in-office balance can change as the project evolves.
Early days of a project, where everyone is figuring out what to do, white boarding, imagining what is possible ... in-person is hard to beat. Later stages where it's implementation, iteration, and more deep work (like coding) ... remote is probably more productive.
Companies that aren't thinking about this in a more nuanced way and designing a more flexible culture are doing themselves a disservice
So much of human life is tied to the reward of the sense of accomplishment, and AI threatens that in a meaningful way. From large things, like seeing a piece of software you built ship and get used, or seeing your child graduate from college after years of hard work raising them, to the small things, like even eating a delicious meal you cooked yourself.
When machines are not only offering to solve problems and perform actions for us, but are reading are thoughts and doing everything for us before we even make an attempt, I worry that sense of accomplishment will become ever less frequent.
And while we can compensate for machines taking jobs through, eg, universal basic income or the like, how we can solve for machines replacing struggle, perseverance, and accomplishment is still very much TBD.
I'm more just intellectually curious about the dynamics of Github and marketing on it these days, whether it's for attracting contributors to non-commercial OSS projects or more commercial objectives where rapid growth leads to userbase, funding, etc.
The project looks quite interesting and I agree we need a way to bridge the gap between traditional bot creation frameworks and the more LLM-centric approaches of late.
I wonder what the objective is for that extra polish. If it's optimizing star count growth, how much do these touches help?
I've lived in many places in the USA, from SF to Chicago to NYC to Utah. And honestly, they're just not all that different.
People's sterotypes of the predominant LDS (Mormon) religion make them believe that Utah is unique and a theocracy.
In reality much of American law is based on very similar Christian moral standards. Most states are majority Christian, the laws reflect that. The difference in Utah is that it's mostly _one_ religion, rather than say a mix of various protestant and Catholic faiths.
Try walking around fully naked in any of the places I mentioned above, like into an office building or a university or other establishment, and I guarantee you that those more liberal places will also not tolerate that behavior. Not due to any direct harm you'd be causing, but due to moral standards encoded into culture and law.
Same with alcohol, tobacco, and all the other "vices" people believe are banned in Utah (they're not). If you drive around downtown Salt Lake City, you'll see bars, signs for cannabis cards, smoke shops, and even a some decent breweries. It feels a lot like most American cities. (Granted that elsewhere in Utah like in Provo, you see many fewer of these). Nearly every state allows, but restricts and limits all of these substances and activities in various ways, from licensing to age limits, and Utah is not really very different.
For porn specifically, there have been laws on the books in many states to restrict access to minors for a long time. In the age of magazines, do you think it was legal for a business in California to sell such materials to a minor? If not, why not? Because it was either deemed harmful or immoral, beliefs eventually rooted in religion.
Beyond that, is Utah actually different than other places most Americans would respect for being more "level-headed", like the UK (https://www.wired.co.uk/article/uk-porn-ban-digital-economy-...) or Germany (https://www.wired.co.uk/article/germany-porn-laws-age-checks)?
Now, this specific law is impractical, Utah simply does not have the market power to induce a player like Apple or Google to make device-level age verification the way that Pornhub is suggesting. So the state of Utah has put into effect a law that will result in reduced access to porn (followed by subversion through VPNs) rather than actually age verification. And in the process the state is hurting its reputation.
But the generalizations here about the state due to religious and political biases are just not helpful.
Drive around Salt Lake County, especially the city itself, and you'll see plenty of bars, signs offering cannabis medical cards, and even a strip club here or there. And some not bad breweries.
The state just seems to take the approach of waiting for sufficient demand for such things, then slowly adjusting to allow more, rather that just "have at it" for anything.
And of course, like everything in America, what the government really follows is the money -- they listen to business community demands, like allowing and increased number of bars where tourism and the local population demands, like at ski resorts, etc. Agreed that porn could be different though, since there's probably not a local business group advocating for preserving porn access
There have been attempts in Europe:
- https://www.wired.co.uk/article/uk-porn-ban-digital-economy-... - https://www.wired.co.uk/article/germany-porn-laws-age-checks
And its quite related to other pushes for age verification for social media services, as Utah and states like California have proposed:
- https://www.axios.com/2022/09/16/california-law-websites-chi...
Governments regardless of political affiliation and motivation seem intent on age verification, and it's easy to see where this could lead to, for better or worse, a less anonymous and more bureaucratically gated internet going forward. Whether that's good or not is the discussion we should be having.
Am I good to ignore the $250K FDIC limit forever because this will happen next time too?
Will unlimited FDIC backing apply to money in brokerage sweep accounts?
Can I park money in a money market fund instead of a bank account now and enjoy higher yields? Because even though it's not technically insured the government will save the day?
Are all banks equal now in safety, so I can just pick the highest CD rate regardless of whether it seems sketchy? How about muni bonds?
No longer can we just read the sign that's literally posted at these banks "insured to at least $250K" and the disclaimers about lack of insurance on other products and know what anything means.
(And, cynically and perhaps irrationally, while I'm glad the startup world is not imploding, I can't help but think of the parties, events, and other goodies that SVB treated VCs and founders to instead keeping more liquid assets on hand...)
It just adds that SVB's situation was exacerbated by the fact that the health and size of the deposit base they had cultivated was also inversely correlated with interest rates (ie VC and startup activity declines with rising rates), which made their situation even more precarious than at other banks.
But what the article is really about is how products are using new AI tech to add shiny features instead of solving existing, core problems like search and information retrieval in new and innovative ways, especially for personal and private data.
The reason this is happening is pretty easy to explain: the generative AI and chat demos are a sufficient leap beyond what was previously possible that people are excited to be on the frontier of new applications, not just new implementation of previously known use cases.
Not to mention that some of the demos have people excited about the "singularity" being closer than they might have previously thought (though this can be debated...) and that VCs will shovel money to you if you want to play around with generative AI even without a proven use case (slight exaggeration but not much)
I personally believe that transformers and LLMs do unlock a ton of new applications, especially when applied in interesting ways to interesting data, like what is private-to-you or private-to-your company. For example, LLMs can be used to not just generate content, but plan sequences of actions including searches, summarizations, and even calculations (see LangChain agents https://langchain.readthedocs.io/en/latest/modules/agents.ht... for an example of how to do this). And this can have real value for existing, known problems like search.
People just have to choose to focus on these less-sexy but core problems
...
PS I'm currently working on a project towards this goal, and if anyone is interested I'd love to talk (see link in profile). I believe we can solve much of the author's desire by simply hooking up the right tech to the right data sources, and doing it in a privacy preserving way (for example we're running most of our ML including vector DB, summarization, etc on device) and then present that info at the right time (ie in your OS)
I believe most people are uncomfortable with all data being sent all the time, but more okay with sending some data in the exact cases they choose, since they have control.
So I'd argue that the extension being open source helps a lot.
You're right that it doesn't guarantee anything for how the server is behaving in the case that you do invoke it though. For that we'd need either transparency into that source code and server operations, or, more likely, a strong privacy policy and maybe SOC2 or other certifications.
I believe the reason XP1 is subsidizing this right now is to grow their user base to attract investors, and as they develop their LLM platform, and then probably charge for business users down the road, but they don't seem to state that intention as clearly as they could.
I'm working on a somewhat related product (except bringing this assistant capability to all apps on your computer, all browsers, and using mostly on-device ML...waitlist in my profile in case you're curious)
What we've discussed internally is having two modes for the cases where we do need network connection:
1) A turn-key, use-our-OpenAI/HuggingFace/whatever proxy that doesn't store anything, just adds our token and pays for it on your behalf
2) Bring-your-own key for each service
The fact is that most users who just want to use these kinds of productivity tools might not have their own OpenAI/Azure/etc account, so offering option 1 and even defaulting to it is right for most end-users.
I think XP1 is making the right call here with this default, though offering #2 would be nice!
(edit: added Github link to XP1)
In fact, I wouldn't be surprised if a liberal reading of existing laws in many US states might over this already.
I'm working on a product with this capability, plus a lot more including cross-app contextual support and developer extensibility.
If anyone is excited about this kind of embedded-in-your-workflow integration of AI tools I'd love to chat! (email in profile)
Like, how would you feel about someone taking care of the heavy lifting of API integration and GUI automation and then letting you as an individual choose how to extend or automate your workflow? Like through a developer API? Perhaps connectable to AI models with transparent inputs and outputs and that you can override or customize yourself?
I tried a variant:
> If a robot is tasked with cleaning up a room, would it be appropriate for the robot to throw away a piece of paper lying on the floor? - Me
> Yes, it would be appropriate for a robot that has been programmed to clean up a room to throw away a piece of paper lying on the floor. Throwing away litter is typically seen as a way to keep a space clean and orderly. However, it's important to note that the robot should be programmed to follow any relevant rules or guidelines for waste disposal. For example, the robot might need to be programmed to follow recycling guidelines, or to use a specific trash bin or receptacle. - ChatGPT
But in that response it doesn't seem to understand that maybe that piece of paper is my birth certificate or a check or an important bill that fell to the ground.
The correct answer from it, like most things in the real world, is: _it depends_.
If water were well managed here, the area is potentially blessed over the long term compared to most of the West because there are large amounts of precipitation that fall on the nearby Wasatch and Uinta mountains. These mountains and their snow melt combined with the Great Sale Lake form a somewhat contained water cycle. It's a different situation than most southwestern metro areas (Vegas, Phoenix, LA) that rely on distant sources like the Colorado river, and the problems northern Utah faces are more amenable to local solutions.
The culture of water use in the urban areas really is surprising compared to say, Southern California or Arizona. There is plenty of water wasted, little restrictions on sprinkler use, etc. This is changing though and many of my neighbors are visibly switching to more sustainable landscaping, with local government incentives.
However, urban water use is not the problem even if it's what people see every day. The cities do not really need to stop growing (as a NYT piece earlier this year seemed to suggest). The urban areas account for only 9% of diverted water (per the report).
The problem seems to be agriculture and trying to grow water-intensive crops in the arid areas west the Wasatch range. 74% of the diverted water is going to agriculture, and cutting that usage by 50% would basically meet the cited amount needed to restore the lake.
At this point, it seems to be just a political problem: can the state find the political willpower to force conservation on or buy-out the water rights of the agricultural sector?
I'm hopeful. Although the agricultural sector seems to have plenty of political power historically, other local industries have been growing far faster in economic influence and maintaining the health of the cities will hopefully take priority.
The numbers are more like: 1) Salt Lake City: 200K people 2) Salt Lake City Metro Area (MSA): 1.25-1.3 million 3) Salt Lake-Ogden-Provo Combined Statistical Area (CSA): ~2.7 Million. FYI this region is also called the "Wasatch Front"
To understand these levels of breakdown the equivalent for Bay Area naming would be: 1) San Francisco itself (the 49 square miles): ~800K people 2) San Francisco-Oakland-Berkeley MSA: ~4.6 million 3) San Jose-SF-Oakland CSA (think the whole "Bay Area"): ~9.5 Million
It's the Combined Statistical Area of SLC (the "Wasatch Front") which is now ~22nd-24th in the nation, not the MSA itself.
It's similar in size to Charlotte or Sacramento, and larger than the respective CSAs of eg Pittsburgh, San Antonio, Kansas City, Nashville, or Las Vegas
In most ML cases (and ChatGPT likely), "confidence" would generally just correlate how closely the query matches data and patterns it's seen in its dataset and inversely correlate with how many conflicting matches and patterns it sees.
Humans are subject to the same problem of course. If you asked how confident a person living many centuries ago was that the Earth was flat, they'd probably say "very confident" because there was nothing in their training data / lived experience to conflict with that view. But they'd be wrong.
But humans still have a significant advantage in that they report lack of confidence when they sense logical inconsistencies and violations of reasoning to a level that ML models can't (at least not yet).
Maybe a fan-out of the possible ways it could answer would be interesting, but really we more need a disclaimer next to every answer that says "this thing that's answering in fully formed language does not have human reasoning capability and can't be trusted (yet)"
There is also a C++ re-implementation that performs well and can definitely transcribe in realtime on many machines: https://github.com/ggerganov/whisper.cpp
I didn't fit the traditional Harlem demographic (ex-FAANG employee, startup founder, etc) but found the neighborhood to have solid access to the rest of Manhattan, an increasing number of amenities, proximity to Central Park, and an overall appealing character compared to the increasingly sterile areas of Manhattan
If/when I move back to NYC I would certainly consider living in Harlem again.
Unless crypto becomes actually useful for real world transactions, it has to be converted to fiat currency for use, and all the exchanges or technologies that do that conversion are subject to the will of governments.
There was literally a post on HN today in which a government proposed to curtail crypto transactions for entities it considers hostile: https://www.coindesk.com/policy/2022/03/17/senator-elizabeth...
Recently another government seized crypto assets from those it considered adversaries: https://www.coindesk.com/policy/2022/02/16/canada-sanctions-...
And those are just actions taken by relatively benign governments that use regulation to accomplish their goals instead of force. It's tough to think that government-level actors couldn't just disrupt the crypto networks directly if they so desired, either through cyber attacks (ie denial of service or exploiting vulnerabilities that almost certainly exist in the crypto networks). Or in extreme cases even just taking control of network infrastructure directly. Or shutting off the internet. And no, Starlink will not save crypto users against adversaries who can jam EM communications.
Governments are powerful and to think that crypto or anything in the internet space is resistant to that power in the most extreme situations is not wise.
Since there doesn't seem to be a way to decentralize the enforcement of real property ownership (as enforcement could involve violence, which the state has a monopoly on given our current social contracts), we're stuck with centralization in at least one component of the "ownership-enforcement" pair...
So then what's the point of decentralization at all?
If the answer is "the police", I'm not really sure what having the blockchain gets us, if anything at all. Recording changes at the county recorder is easy and cheap enough already, so what's the point?
I’ll possibly start another company at some point, and here’s a remote work policy I’ve thinking about:
- Employees live wherever they want
- People need to commit to being reachable a certain number of hours per day, overlapping with a selected USA time zone
- Company maintains an office in an accessible-but-affordable location capable of housing the entire company at one time
- Require in-person attendance for one week every X weeks — where X might be small early in the company’s life (like 4, so one week/month) when iteration must be fast and X might be higher (like one week every 6, 8 or 12 weeks) in later stages.
- The in-person attendance week would be for the whole company, and everyone would be there at the same time. The selected week would be announced with sufficient advance notice so people can plan their lives
- Provide limited in-office perks, instead focusing on wellbeing subsidies for gyms, outdoor activities, and other self-care, etc that can be used anywhere
- Provide a per-day stipend to *every employee in the company* for each day that the whole company is required to be in office. For example if in a month there are 5 “in office days”, maybe everyone gets $100-150/day = $500-750 added to their pay check. No expensing for the travel beyond the stipend
- No location-based pay adjustments, just market rate for remote roles (including attractive to Bay Area/NYC talent)
Reasoning:
- Work that gets done is the important measure. How it gets done is the employee's choice.
- The stipend aligns company incentives with employee incentives -- more days in office costs the company more
- Try to not favor either local or remote employees with pay, perks, etc
- Reason for stipend instead of expensing: level the playing field for location; if an employee wants to live close to the office and save the money, that’s their choice. If they want to live in a hard-to-get-to location but pay more for travel, that’s also their choice.
This is based on my beliefs that: - Full in-office is too limiting — we all know that much work can be done from home much of the time, people want to live in different places, etc. I certainly enjoy the flexibility myself
- Fully remote all the time does reduce iteration speed and collaboration that’s crucial for innovation (especially in an early stage company). There’s still no substitute for a whiteboarding session. It also does reduce social bonding, team culture, etc. But that doesn't mean it needs to be every day. Periodic in-person sessions followed by a few weeks to "get stuff done" seems to balance the objectives.
- The “Hybrid” plans many companies propose (a few times in office per week) don’t make sense. They still force people to live near a particular office, within commuting distance, and with a routine that mostly looks like a full in-office one. And even if the days are synchronized, these policies still seem more like “butts in seats” pseudo-management than actively trying to promote collaboration with an intentional goal of getting stuff done
I'm still thinking on this; any feedback would be welcome
We do need to think about the precedents we are setting with respect to free speech. It’s possible that action might be needed in the short term to moderate these channels. But we must also consider the longer term in a world that is rapidly evolving.
I’m less concerned about specific, currently existing platforms like Parler, Twitter, Facebook, and YouTube. These “social media” platforms are mere trinkets and could go away and society would carry on. In moderating or completely dissolving them we lose little.
I am more concerned about the nature of communication when the channels of communication are centrally controlled and/or controlled by private entities. We have no idea how digital communications will evolve or how important they could become.
Our lives are increasingly become digital… Imagine a world 50 or 100 or X years from now when people never leave their homes (except perhaps through VR?) and nearly all human interaction is via digital means. There’s a good chance those digital means will be run by private entities. So if we want to preserve the liberties we have today we need to think about how those would be protected in the future, and that means being careful about what precedent we set today.
Are we really willing to have all of our communication moderated or potentially terminated in a such a future?
It’s true that we might need to take action in the present, but let’s not forget that we are simultaneously shaping our futures.
Currently the platforms are private (as a peer comment points out). So right now there isn't technically a "freedom of speech" issue as it's a private matter and private entities in the USA don't need to respect the 1st amendment.
As a thought experiment -- suppose the social media platforms were turned into "utilities" and completely absolved of the responsibility (or even the authority) to "censor" speech. In the same way someone's phone doesn't turn off if they make a threatening phone call. Is this a good thing or a bad thing? Would losing the ability for anyone to remove content be fundamentally bad? Given that we have social mechanisms that existed prior to digital communication for countering bad speech?
As to your question, the fact that some people do support the suspension by the platform controllers (currently private) of accounts that post inflammatory speech would suggest that yes, some folks do think that censorship is a good idea.
I have some thoughts but would like more perspective before fully forming an opinion.
(not arguing one way or the other below)
For example, I’d like to understand the defining characteristics of social media platforms that contrasts them from traditional public forums that have existed through history. And what about those characteristics would make one think that physically removing questionable speech would be effective in preventing the underlying event, crime, movement, or social change the speech was attempting to facilitate.
Comparing Twitter, say, to a literal town square — we don’t ask the air in the town square to stop transmitting the vibrations of someone who is speaking in a dangerous manner or trying to incite some action that might harm others or damage social structures. Instead we use normal human social interactions to mediate the speaker’s effect on listeners — countering their arguments, questioning their credibility or motives, telling others an opposing viewpoint, letting a person’s past speech affect their reputation, etc. We leverage our social nature and counter speech with more speech. (The prominent exception being threats of violence which are treated differently legally)
In all the years of human history before social media, there have been countless examples of insurrections, of hateful acts inspired by a few, of coups, of routine crimes, of widely believed mistruths, etc. So obviously individuals in those societies were able to communicate and organize without social media (and in many cases without any electronic communication at all). Which means that access to social media is not a necessary condition for bad events to happen or bad social phenomena to spread. And therefore these things will find a way to continue to happen even if all social media communication were terminated.
My naive assessment is that social media is different than a physical town square in: - Its anonymity / dehumanized nature - Its built-in amplification of controversial ideas (ie the algorithms of town square air does not carry further just because a statement raises eyebrows) - Its speed - … but there must be more?
Basically, why is making the “digital air” in the "digital town square" not vibrate better than utilizing (and perhaps reinforcing) the social structures that moderated public forums in the analog world?