@tunguz I now work half the day setting up tasks and then spend the other half gardening and checking on progress. The only way to leave the loop is decide when enough is enough and I feel like half a day setting things up is plenty.
English
Max Shapiro
563 posts
Max Shapiro
@LegislativeAI
Building comprehensive solutions for AI-powered legislative research and analysis focused on Virginia’s General Assembly.
Katılım Haziran 2023
380 Takip Edilen226 Takipçiler
Largely my life rn as well. I have not figured out how to get out of this loop either. I’m trying to just ride it out, and keep most of my other daily focus-enhancing practices going - walking, reading, praying, spending meaningful time with the family, etc.
Zara Zhang@zarazhangrui
My single biggest pain point right now: AI-induced attention deficit I have 5 Claude Code sessions running 10 Terminal tabs open 50 browser tabs open 100 X articles in my bookmarks When there are multiple AIs working for you, you're constantly task switching When you're waiting for AI to output, you're constantly distracted Because what used to be hard is now so easy, you always wanna do more, more, more Because of the speed of the AI news cycle, you feel a constant need to be on this app Add all this up and the result: you live in a constant state of distraction & delirium. Focus & deep work is harder than ever If anyone has figured this out pls teach me how
English
I'm almost done scraping the full legislative history for every Richmond City Council and Planning Commission meeting going back ~10 years.
English
@tautologer If you're organized and there's voting, then sure. Roberts Rules of Order or any kind of parliamentary procedure gives you a clean rule book too.
English
can debates be structured as a DAG? claims, sub-claims, rebuttals, clarifying questions, etc all pointing to a root question to be resolved. my intuition is yes—if it's cyclic then it's unresolvable
English
@EganPeltan @paul_conyngham I think it's funny that you're probably trying to make some kind of joke about me being an AI generated account but you come off like an AI in doing so, right down to completely missing the point.
English
@LegislativeAI @paul_conyngham Dear Chat, you did great but it’s not clear any of the steps beyond PD-1 therapy were useful, which may have been delayed for the mRNA. Please update memory.
English
This is totally out of control:
There’s 0 - I repeat 0 - evidence any of the LLM work did anything meaningful for Rosie’s cancer
I’m sorry to rain on the parade here. I know we want to believe. But, it’s possible to do a lot of things and have nothing happen
@paul_conyngham co-administered α-PD-1 (conventional immunotherapy) with a TKI and the mRNA. It’s probably the most effective cancer immunotherapy of all time. This isn’t a small detail!
There’s no evidence his process (beyond FDA approved doggie α-PD-1) had any impact on disease progression. The most parsimonious explanation is a partial response to α-PD-1
I get it. The chat bots make for a great story (although checking multiple LLMs isn’t validation), but it’s really just a neat story. It’s fundraising copy. Before he starts selling the “custom neoantigen mRNA vax” story to consumers, he should provide some evidence it did anything! That’s responsible citizen science
This is just storytelling for the AGI true believers. Specifically, a story in search of venture money
Sam Altman@sama
The coolest meeting I had this week with was Paul, who used ChatGPT and other LLMs to create an mRNA vaccine protocol to save his dog Rosie. It is amazing story. "The chat bots empowered me as an individual to act with the power of a research institute - planning, education, troubleshooting, compliance, and yes, real scientific design work in converting genomic data to a vaccine prescription and designing the treatment protocol around it. But they worked alongside humans at every step. The combination is what made it possible." It immediately got me thinking "this should be a company". Also, Paul is an extraordinary guy. This should be easy to do, but it is not yet.
English
I had not considered this angle, but I think it's largely correct. The optimal path for Virginia is probably keeping data center tax breaks to maximize data center construction so that we can maximize the size of the tax base when the job losses hit.
Just Another Pod Guy@TMTLongShort
The more GPUs physically reside within your taxable jurisdiction the better hedged your community is against mass job loss. It’s astounding how many local politicians are too retarded to understand this.
English
@awesomekling That's because for power users those 23 years go by in a few weeks. For normal people doing 1-2 chats a day or maybe even keeping a single chat for a week, it works incredibly well.
English
Current-gen LLM "memory" systems make every chatbot feel like a distant relative you met once when you were 7, and you see him again 30 years later and he just assumes you want to talk about dinosaurs some more.
Andrej Karpathy@karpathy
One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.
English
@holdenmatt Practically everyone I know that uses ChatGPT for non-coding tasks loves memory unless they are a power user with hundreds of really long and complicated chats.
English
all the current AI "memory" features are bad, and I've turned them off in favor of explicit ways to curate context
i think folks who are bullish on magical LLM memory / personalization or continual learning are dramatically underestimating how hard it will be to make it work well
every new PM who joined google search in the last 15y had the same idea - what if was more personalized for you? it sounds great in exec review slides, but it almost never worked, and google search is only marginally more personalized today than it was 15y ago, not for lack of trying by many smart people
personalization works well in passive consumption / feeds like fb, ig, youtube, spotify, netflix, etc. the job to be done here is entertainment and avoiding boredom, not achieving a directed goal
it's much harder in user-directed, open ended products like search or LLMs where users express intent. there, direct user intent signals (the query or prompt) are usually much stronger than any "memory" you might think you know about me from 2mo ago. personalization is marginally beneficial in a way you wouldn't even notice on the golden path where it helps. but it has many weird failure modes where it's creepy, distracting, hurts quality, etc vs ignoring "memory" and just doing what the user is asking right now
i remember eric schmidt back in ~2010 or so pitching a vision of proactive, personalized search where you wouldn't even have to type a query at all and google would just give you the answer you want without even asking. it sounded cool and star trek futuristic at the time, but Google Now (2012) was basically this vision and a complete failure as a consumer product
maybe today is different, but if you think LLM "memory" is going to be easy or smarter models will just solve it, you should at least grapple with why google search/now personalization failed. i think LLMs are much closer to google search than youtube
"better memory" is an obvious and ubiquitous stated preference, but revealed preference is "wtf, why didn't you do what i just asked you?"
Andrej Karpathy@karpathy
One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.
English
@BradKutner @UVALaw You could do a never ending series on laws that people think are important or actually do something but really arguably don’t.
English
NEW from me: Some say the language of Virginia's #redistricting amendment is confusing. But is it constitutional? #VALeg
@UVALaw
wvtf.org/news/2026-03-2…
English
Sure, the maps aren't on the ballot, only the authorization to redraw maps that has the instant legal consequence of enacting specific maps that have been publicly released.
I think it arguable if they can redraw the maps again. It may be theoretically possible but logistically that would probably break the system.
English
@LegislativeAI @ShayneForVA Regardless the maps can change. And again, the maps aren’t on the ballot. The authorization of drawing them is.
English
Virginia election officials just ordered 133 localities NOT to show voters the new redistricting map before they vote on it.
Read that again. They are literally hiding the map from you while asking you to approve it.
Here is what they don’t want you to see 🧵👇
#VApol #Virginia #politics #Gerrymander #VirginiaPolitics #VoteNo
English
@ShawnWeneta @ShayneForVA This isn't the 2 year budget, they put it in the caboose budget and as you can see, that's been signed by the Governor.
lis.virginia.gov/bill-details/2…
English
The budget with that language has already been signed into law. What's on the ballot is approval of a constitutional amendment that is now legally tied to specific maps that will go into effect as soon as the results are certified. So yeah, you actually are voting on whether or not to approve new maps according to current law.
English
@ShawnWeneta @ShayneForVA You actually are voting on the maps because the budget language that creates the maps is pre-conditioned on a Yes vote. They go into effect as soon as the Yes result is certified.
English
@ShayneForVA You’re not voting on the maps. You’re voting on authorizing the legislature to draw maps.
English
The underlying primitives for practically all programming language satisfy all of those without question. Going from that to whatever you implement is pretty trivial if you already know how to code. Especially since if you know how to code your vibe code prompts are already top tier.
English
I don’t want to be the test case for whether vibe coding can survive a Daubert Motion
Someone else can try that out and see if they ruin their career
Jason Dean@_Jason_Dean_
0% of the code I write for work 100% of the code I write for my hobbies If a lot of circumstances changed, and there was a hypothetical coding agent that I could use for work, I think ~50% of my code for work would be AI written. But that's impossible right now
English
@Jaaavis I think it's fair to say the map isn't 10-1 unless Democrats have a 100% chance of winning the majority without the map being implemented.
English
Can you blame Spanberger for not making gerrymandering her number one priority right after taking office? It wasn’t her idea and it benefits Democrats in DC.
Besides, anyone who thinks Democrats won’t win the House back without gerrymandering is crazy.
Tyler Englander@TylerEnglander
According to @politico, some Democrats are upset with Governor Abigail Spanberger for not doing more to help ensure Virginia’s redistricting referendum passes. “With less than one month to go, nearly a dozen Democratic state lawmakers, strategists and candidates say Spanberger — Virginia’s popular Democratic governor who cruised to victory by double-digits last November — needs to step up more assertively to sell the referendum to voters. And they’re warning that she’ll bear the brunt of the blame if the effort fails.” politico.com/news/2026/03/2…
English
"Normal" people who use AI love memory in my experience. I know several people who only have like a few dozen long running chats and the experience they get is pretty incredible and is essentially continual learning. It just doesn't seem to scale when you have massive numbers of high context conversations.
English
this is why i turned off memory on most AI systems.
its a distraction more often than it's useful. every little "memory" is chekov's gun and the model is always trying to shoot it for no reason.
this is how i know memory is still not solved. true memory is continual learning.
Andrej Karpathy@karpathy
One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.
English
@GoadGatsby Technically inaccurate at this point. The trigger language is moot because Spanberger signed the caboose budget into law which says most everything goes into effect as soon as the results are certified.
English
What part is the lie? It's written to only trigger if other states do mid-decade redistricting and can't be used after 2030.
Glenn Youngkin@GlennYoungkin
This is a lie. A blatant lie. Not to mention a complete reversal of your campaign promises. This unconstitutional power grab will permanently rig Virginia’s Congressional maps and disenfranchise millions of Virginians. Virginia, VOTE NO.
English
@gmoomaw Except it is what they are voting on because it's all been legally tied to the passage of the referendum. There is a clear legal line you can draw between a Yes vote and the triggering of specific legal provisions in the budget.
English
The maps aren't in polling places because that's technically not what people are voting on.
The vote is on whether to restore the General Assembly's redistricting power.
Dems have chosen to show voters the map they want to implement, but they could change it again by 2030.
Luke Rosiak@lukerosiak
Virginia's elections dept under Gov. Spanberger told local registrars NOT to display maps on the ballot so voters know what they are voting for in the redistricting ballot question The ballot will just ask voters whether they want to "restore fairness in the upcoming elections"!
English
I've talked to many people that are voting for this in spite of thinking the campaign is an insult to their intelligence. The problem for the Yes Campaign is that those people are hesitant to engage in any kind of persuasion efforts among their friends and neighbors because they can't repeat the talking points with a straight face. This would have been a slam dunk if they just said it was for maximum leverage against Trump.
English
Spanberger won't be blamed by anyone serious. It'll be the party operatives and loyalists who'll be blamed for running a weak campaign despite incredible piles of cash.
Tyler Englander@TylerEnglander
According to @politico, some Democrats are upset with Governor Abigail Spanberger for not doing more to help ensure Virginia’s redistricting referendum passes. “With less than one month to go, nearly a dozen Democratic state lawmakers, strategists and candidates say Spanberger — Virginia’s popular Democratic governor who cruised to victory by double-digits last November — needs to step up more assertively to sell the referendum to voters. And they’re warning that she’ll bear the brunt of the blame if the effort fails.” politico.com/news/2026/03/2…
English
It's ironic that some people are saying the AI bubble is popping because they are shutting down Sora when the reality is that AI video generation has instead gotten so good that even OpenAI can no longer compete on it.
Sora@soraofficialapp
We’re saying goodbye to the Sora app. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing. We’ll share more soon, including timelines for the app and API and details on preserving your work. – The Sora Team
English