getpost

Ken McLeod[0] is teaching a 6 week class Tuesday evenings on the Diamond Sutra at The Alembic[1][2] in Berkeley. The first class was last week. It may be possible to join now and watch the first class online.

Ken recommends the Red Pine[3] and Thich Nhat Hanh[4] translations.

[0] https://unfetteredmind.org/ [1] http://berkeleyalembic.org/ [2] https://www.eventbrite.com/cc/practicing-the-diamond-sutra-w... [3] https://www.amazon.com/Diamond-Sutra-Red-Pine/dp/1582432562/ [4] https://www.amazon.com/Diamond-That-Cuts-Through-Illusion/dp...

> Don't be paranoid.

How do you mean? I think the article (and my experience) suggests that you do have to be paranoid. [I looked up paranoid, just to be sure I knew the exact definition, and I didn't. It's an "extreme and irrational" fear. Is looking it up parnoid? Hahaha.] Colloquially, paranoia is extreme and not necessarily irrational. Think of Andy Grove, "Only the paranoid survive." Or Kurt Kobain, "Just because you are paranoid, it doesn't mean they're not after you."

Anyway, the way I frame the issue of software quality, is to hold the view that there are always errors, and the best you can do to apply extreme vigilance in attempting to ensure errors occur rarely.

GPT4o access is a handy feature, but, what I was hoping to hear about is an improvement in Siri's language "understanding."

In today's WWDC presentation, there were a few small examples of Siri improvements, such as an ability to maintain context, e.g., 'Add her flight arrival time to my calendar,' wherein Siri knows who "her" refers to.

In my day-to-day experience with Siri, it's clear Siri doesn't have the kind of ability to understand language that LLMs provide. It still feels like clever son-of-Eliza hacks with stock phrases. If your utterance doesn't match with a pre-programmed stock phrase, it doesn't work. The other day I said something like "Play the song you played before the one I asked you to skip," and Siri didn't seem to know what I wanted. OTOH, GPT4o can easily handle statements like that.

Does anyone know to what extent Siri's underlying language models are being upgraded?

I think ketamine is niche, and it will remain so. It's one of many tools. MDMA also assists in restoring brain plasticity[0].

For me, by far the most interesting development in mental health is Chris Palmer's work, which asserts that most mental illness is primarily a metabolic issue. If you think that can't be possible, read his 2022 book[1]. Here's a podcast[2].

[0] https://www.nature.com/articles/d41586-022-02871-w [1] https://www.amazon.com/Brain-Energy-Revolutionary-Understand... [2] https://www.youtube.com/watch?v=oC-sQogfh3Q

Several adults in my friendship circle (retired or semi-retired) have evolved to spending nearly their entire waking lives online. They're able socialize normally, but they don't make the time to do that as often as in the past. This is tantamount to hikkormoridom.

One friend went to visit two other friends who live together in New Mexico. He imagined they'd be out and about doing stuff during his visit, but the hosts remained preoccupied by their online activities. The visitor could have stayed home and texted.

I had a IIcx with a Radius FPD, a dream machine! I could see a "page" of code without scrolling. I'd forgotten about the Apple FPD.

It begs the question, is outlier success mainly due to crazy ideas combined with a lack of discernment?

No pricing info, at least not without signing-up?

Good luck building apartments in active mudslide areas. Berkeley residents will be lucky not to lose the single family dwellings that are already built. Slide hazards will only get worse with ongoing climate change. See the map at the following link!

https://www.berkeleyside.org/2023/01/20/risk-of-major-landsl...

Storing data on a VHS tape was a thing in the early 1980s. I worked on an AlphaMicro "minicomputer" with this feature.

https://manualzz.com/doc/o/9paan/amos-system-commands-refere...

Nice! Two episodes on Ryoan-ji https://www.youtube.com/watch?v=ldVz7XQFky0 https://www.youtube.com/watch?v=9vphfpgGL7A

In my startup experience (in the 1980s) as a recent CS graduate, there were 2 disheartening situations:

1) An intern was paid 25% more than me. The disparity was even greater, since he was paid on a hourly basis, and I was salaried, working 60 hr+ weeks. It was baffling, the intern wasn't a family relation, and had no special skills. In fact, I had to onboard him. But, for whatever reason, he got an absurdly high wage. I suppose the hiring manager (who later went on to be an early Microsoft VP) wanted to be magnanimous and help a college student, but why not also be magnanimous with your regular employees who are busting their butts?

2) Starting salaries for new grads were increasing faster than salaries for existing employees, and salaries for existing employees weren't adjusted accordingly. Existing employees got stock options on more favorable terms though (having started earlier).

Yes! https://en.wikipedia.org/wiki/KH-11_KENNEN

I first read about this in the 2007 book, Extraordinary Knowing: Science, Skepticism, and the Inexplicable Powers of the Human Mind by Elizabeth Mayer. I don't remember what she says about Swann, but the book made a big impression on me at the time, and I bought copies for a couple friends.

https://www.amazon.com/Extraordinary-Knowing-Science-Skeptic...

I did not know this, or forgot, thank you for saying! There's applause and laughter after the short version of the quote, so it struck a nerve, and not without reason.

There are comments on Smalltalk at various places in the talk, but after the quote, all he says is, ..."There is one really important thing about Smalltalk, and some of languages like it, that we should pay really really close attention to, but it has almost nothing to do with the syntax and accumulated superclass library," ....

Quote is here: https://www.youtube.com/watch?v=oKg1hTOQXoY&t=634s

I'll see you Bjarne Stroustrup, and raise you Alan Kay,

"Actually I made up the term "object-oriented", and I can tell you I did not have C++ in mind."

https://en.wikiquote.org/wiki/Alan_Kay

> the researchers say that children may be "evolutionarily primed" to expect exceptionally high levels of physical contact and care,

What strikes me here is the phrase, "exceptionally high levels." I imagine children need a "normal" level of care; it's only "exceptionally high" in comparison to the deprived state of family systems in these degenerate times.

I often reflect on the understanding in Attachment Theory, wherein a child a needs to have a caregiver who is sufficiently attuned to the child's needs. And it turns out, "sufficiently attuned" means that the caregiver responds in an attuned manner to 30%-50% the child's entreaties. As one of my meditation teachers says, 'That's not a high bar. What grade did you get the last time you scored 50% on a test?'

I see lots of complaints about street names not rendering, and that has been a problem for me as well. The tweet author has a separate thread about how India doesn't routinely have street names, and how that lead to a design change for maps. It's super interesting!

https://twitter.com/elizlaraki/status/1719379064564166772

EDIT: The following link is among the replies in that Twitter thread.

Anatomy of an Indian Address [50th Edition] [AUG 7, 2023] https://newsletter.theindianotes.com/p/anatomy-of-an-indian-...

I asked for a list of Federal Chancellors of the Federal Republic of Germany and GPT4 included Scheel ("acting"), which makes Schröder the 8th. I often troubleshoot GPT4 replies by asking it for more context. "Is" vs "was" is interesting an find!

https://chat.openai.com/share/ddd2800e-630c-4ff7-9992-14b265...

In case you're not already aware, GPT4 et al write usable Python and can execute some Python code themselves for trying things out.

I find it much more pleasant to ask GPT4's advice and have it write sample code than it is to use web search, Stack Overflow, etc. Even when I know exactly what code to write, it's often faster to ask GPT4 to write it a certain way and then make minimal edits myself.

The paid version of GPT4 used to be the best, but lately the VSCode Insider GitHub Copilot produces comparable or identical results, since it uses GPT4. I have only one custom instruction, "If writing Python code, always use context managers where appropriate."

EDIT: If you don't want to hassle with maintaining a development environment, try Replit[0]. Their AI is not nearly as good as GPT4 though.

[0] https://replit.com/

Right, humans are involved in fine-tuning, at least for the time being. GPT4 says, ..."core content creation and verification are human-driven tasks."

I've been wondering about how to do incremental updates without incurring the cost of a full recalculation of the training data. I suppose I assumed that LLM providers would (if not now, eventually) incorporate a fine-tuning step to update a model's self-knowledge before making the model available. This would avoid including the update in the context length.

Among many, many applications, this would be helpful in allowing LLMs to converse about the current version of a website or application. I'd want a sense of time to be maintained, so that the LLM would know, if asked, about various versions. "Before the April 5, 2023 update, this feature was limited to ..., but now ... is supported."

I asked GPT4 about incremental updates, and it seemed to validate by my basic understanding. Here's the conversation so far:

https://chat.openai.com/share/00fe148a-13aa-4e92-8b77-f0de48...

>So, Bard can now access and analyze your Drive, Docs and Gmail!

I asked Bard if I could use it to access gmail, and it said, "As a language model, I am not able to access your Gmail directly." I then asked Bard for a list of extensions, and it listed a Gmail extension as one of the "Google Workspace extensions." How do I activate the Gmail extension? "The Bard for Gmail extension is not currently available for activation."

But, if you click on the puzzle icon in Bard, you can enable the Google Workspace Extensions, which includes gmail.

I asked, "What's the date of the first gmail message I sent?" Reply: "I couldn't find any email threads in your Gmail that indicate the date of the first email you sent," and some recent email messages were listed.

Holy cow! LLMs have been compared to workplace interns, but this particular intern is especially obtuse.

Robert Sapolsky has a new book on free will, Determined[0], and he's doing lots of podcasts[1] about it. I started with Origins[2].

[0] https://www.amazon.com/Determined-Science-Life-without-Free/... [1] https://www.listennotes.com/search/?q=robert%20sapolsky&sort... [2] https://www.listennotes.com/podcasts/the-origins/robert-sapo...

USA v Google exhibit via https://twitter.com/jason_kint/status/1718830641016414657

I enjoyed this 2-Aug-2023 podcast interview covering some of this stuff: https://josephnoelwalker.com/147-katalin-kariko/

tl;dr: The "Small Bets Newsletter" says YC isn't suitable for small bets.

YC isn't designed for small bets! That's not their thing. It would be interesting to see stats on non-huge exits though, some kind of distribution showing the following with a time element, indicating, say, the number of months for various outcomes.

Huge exits 1.25% (define huge), Exit exceeding YC's investment, YC investment repaid, YC investment partially repaid, YC investment entirely lost

I have a Pro account. Not having to install software locally, and being able to do shared online development is very appealing and helpful. To put it another way, Replit saves me from having to spend time learning and monitoring "administrative" aspects of development. I can just work on my coding challenge. Unfortunately, the new billing system is a step backwards.

The problem for me is that hosting costs are now variable, and the mere fact that I have to monitor or worry about an unexpectedly high bill is a drop-dead. Moreover, I have to take time even to understand the billing system.

If I understand the new billing system correctly, and again, I don't want to have to spend time figuring it out, there is a monthly allowance, so that my typical small test deployments are unlikely to incur any charges beyond my prepaid annual subscription. Is that right?

Suppose I leave a site deployed, and there's a DDOS attack? Or by some fluke, the site goes viral because a Reddit user happened to like a test image on the home page. My allowance is exceeded, and I'm suck with a bill. Sure, it's unlikely, but it could happen. I see complaints about unexpected bills from various cloud providers often, nearly every week.

The whole appeal of Replit's earlier billing system is that there was a fixed cost. Is there a setting so a deployment will be taken offline if the prepaid allowance is exceeded? If not, I won't use the service as much as I'd like. Why pay for a service I use only for the occasional one-off?

That said, this makes me realize I'm not a satisfied customer, and I regret renewing my one year plan recently. The site is just too unreliable. A problem the replit has had for years (I haven't tried this recently), for example, is that if I accidentally select a large file in the editor (like a data file), the browser freezes while an attempt is made to render the file. Sometimes the only way to recover from this is to delete the repl. At least put up an alert if a large file is selected, "You have selected a 1 GB file....."

Also, the forum-based support is practically unusable, unless someone happens to have recently posted about the problem I'm having. Searching through posts and reading lot of unsolved, minimally categorized issues with non-authoritative replies is not a good use of my time. I get that Replit doesn't want to provide support, and there might not be a market for paid support in Replit's user base, but Replit can do better. EDIT: Is anyone working on ML-based support?

Made me look it up: https://www.merriam-webster.com/grammar/what-happens-to-name...