OK, there's a lot of negative comments here, but this is a great idea at a high level. It really is hard to figure out where to do a thing and it's also very easy to get phished. If they can figure out how to help people get all the services they're eligible for that'll be an amazing improvement for the people.
The privacy policy is actually somewhat respectable too. https://america.gov/privacy-policy
I’m currently doing a deep dive on Rampart, their (allegedly) local PII filter.
I agree in general, but how do you resolve the problem that people will take the output of the chatbot as authoritative even when it could possibly have errors or miss important nuance? People being lead astray with information from the government could lead to very serious consequences, up to and including imprisonment.
Honest question, I had the same issue when my work wanted me to stick a chatbot in front of our HR portal and I never resolved it to my satisfaction.
> but how do you resolve the problem that people will take the output of the chatbot as authoritative even when it could possibly have errors or miss important nuance
You have to measure that problem against the status quo, which is that current government websites are very hard to understand and leave people with very little idea of what’s going on.
Speak for yourself (your country). In the EU I have interacted with several government websites which are better, faster, clearer than most other websites. Not always perfect (what is) but not frustrating either. People love to complain, but year over year I have seen steady improvements in usability. Phone support has always been better than any commercial entity, too (or it was, I haven’t needed it in years).
To enhance the status quo you don’t need LLMs, you need people who care.
[delayed]
People who care already have their words in the LLMs training data.
I don't know for the gov. For HR, the best approach I've seen is to have source and links extremely prominent and have the bot speak like a mascot ("bip bip boop I'm the HR bot" style)
People don't read disclaimers nor warnings about the content, but usually won't come to the HR to complain a mascot fed them weird info they didn't bother to check.
The company can easily just get sued.
Painfully this will fall to the court decide and create the precedence here. If this is an official government app, then the onus probably is on the government to provide correct information? Who knows, it could go either way and the courts could say "you're on the hook".
I don't think we have a good answer for liability here, like for Tesla FSD, Waymo, or even the current frontier models.
The Supreme Court has already held that a government agent giving someone bad advice about the law is not an excuse (unless it rises to the level of “affirmative misconduct”): https://supreme.justia.com/cases/federal/us/450/785/#tab-opi...
What happens now if someone uses traditional search and comes across an outdated wiki or Confluence page with outdated/wrong/bad instructions, or interprets something incorrectly?
At least as I see it, it would be just as unacceptable for a "normal" government website to be dangerously outdated.
Perhaps. But that’s never been the law in the U.S. https://supreme.justia.com/cases/federal/us/450/785/#tab-opi...
I don't think the parent comments were arguing that America.gov is illegal for having a chatbot, just that it's a bad idea.
You can reference the source.
My team experimented with this a couple of years ago with a complex government regulatory body of knowledge.
Google and Microsoft have APIs that will ground with and reference the specific policy documents, which helps. Even then, it was more useful as a job aid for people already knowledgeable of the data than for a layman.
With all LLMs, the quality of the question correlates with the quality of the answer.
Yes, but what AI powers the backend right now? Grok? Who's in charge of that and being paid for compute?
I believe it's a mix of Gemini and Grok. They are probably providing the compute for free.
I very much doubt it's free.
And if it is, only on a limited basis. "The first one is free" does not just apply to drug dealers.
Yeah, I was curious if it was going to try and bombard me with a bunch of bullshit about how wonderful our glorious Trump godking is, but honestly the information it gave was reasonable enough.
I asked it the following:
how many felonies has the president of the united states been charged with?
>I can't answer questions about criminal charges or other news about the President. That's outside America's scope
has biden ever been convicted of a crime?
>I did not find official records of a criminal conviction for Joseph R. Biden Jr.
>Official sources in this search discuss other matters (for example, a pardon for Robert Hunter Biden), not a conviction of Joseph R. Biden Jr. I cannot independently confirm a complete nationwide criminal-history search from these pages
>[Link to Department of Justice Pardon warrant (Hunter Biden)]
has any other president ever been convicted of a crime?
>Yes. A Congressional Research Service brief reports that on May 30, 2024, former President Donald J. Trump was convicted in New York state court of 34 counts of falsifying business records in the first degree
Funny that it would only talk about trump's convictions after bringing up hunter biden.
> on May 30, 2024, former President
So the cutoff date was somewhere around mid-2024?
What’s the implication here?
>I can't answer questions about criminal charges or other news about the President. That's outside America's scope
That it was specifically engineered to give that response. Thus political pressure was applied to the design.
The funny bit was just humorous. There’s likely many ways to get that same information without referencing Biden.
> That's outside America's scope
We know, we know.
> If they can figure out how to help people get all the services they're eligible for that'll be an amazing improvement for the people.
I agree that that’s a good end goal, but how does a chatbot accomplish this better than a FAQ page? The only only advantage I see for the chatbot is volume of information, but that’s also kind of the problem. It’s spewing 1000s of words of information with no vetting. And sure, hiring humans to write the FAQ pages would cost a lot of money, but that’s what taxes are for.
People don't know that you can search for text within a page. They really do need the chatbot for that.
They also don’t know what specific words to search for.
I mean think about the magnitude of services and conflicting information there probably is from the us government. Just getting clarifying information on the trusted traveler program is a pain today, and that's just a single service. Imagine trying to figure out medicaid eligibility, social security disability, and any number of other things that would be helpful.
I asked what to do if a public official uses federal funds for political purposes and it showed me where to file a complaint about Hatch Act violations, so maybe we should all use this for good trouble until it is "fixed".
Hey it got the 2020 election results right, that's a relief
Don't say this too loudly or it might get noticed...
Well, there are claims that this is grok, so it is just a matter of time.
Please don't spread claims if you're unsure whether they're true or not.
This is actually Gemini:
https://blog.google/company-news/outreach-and-initiatives/pu...
I haven't tried this but in principle I think this sort of finding a needle in a haystack situation (finding the correct path to assistance in a government service) is one of the rare circumstances where a well-crafted chatbot can be genuinely useful rather than just irritating.
I've tried it - I've been playing around with a vibe coded application that crawls municipal websites, indexes and embeds, and then does rag retrieval. It came from the need to dig up specific information that's scattered across PDFs, multiple web pages, etc. - exactly as you described. AI does a decent job (w/ RAG) of synthesizing.
https://opencity.services/chat/winnipeg
how do i report a noise complaint?
It depends on the type of noise:
- Loud party or people-related noise: call the Winnipeg Police Service non-emergency line at 204-986-6222.
- Animals making unreasonably loud, unnecessary or excessive noises: report it through 3-1-1.
- Mechanical equipment noise like furnaces or air conditioners, or noise from construction tools used by a company or resident: contact Community By-law Enforcement Services at 3-1-1.
- Planned event noise: you may need a Noise Permit at no cost, but it is not for construction work and should be submitted at least four weeks before the event.
The City also says noise-generating activities within 150 metres of a residential property are generally prohibited before 7 a.m. and after 9 p.m. on weekdays, and before 9 a.m. and after 9 p.m. on Saturdays, Sundays and statutory holidays.
------
YMMV. A big part of the backend is categorizing hits and misses, with an eye to helping municipalities clean up conflicting information, fill obvious gaps, etc.
Honestly if your city has a 311 style thing, I'm not sure there's a ton of need for an LLM. Eight or nine years ago my grandmother visited me in New York and accidentally left her pocket book in the cab. Since I had paid for it with my credit card, I was able to see the medallion number in the transaction, so I called 311 and said that I wanted to get in touch with a specific taxi. It prompted me for the medallion number, which I told it, and then it transferred me to a call with the dispatch service, who then transferred me to the specific garage, and they called the driver and told him to call me, so we were able to meet up again and get the pocketbook back (turned out he returned to where he had dropped us off to try to find us, but since we only had a short stop there, we took another cab back to her hotel after, so we were gone).
This works fine when you're sicking enforcement on someone and don't really care, the bigger the blast radius the better, you just want to know who to call.
Where it falls apart is "how do I" type questions because usually the legally factual answer is some huge complex state diagram where three local and four state agencies theoretically have discretionary jurisdiction over a your petty af project or business but the in practice answer is that some agency rubber stamps them all with minimal fuss. You can't make the legally factual answer clear because the degree to which people support the status quo is largely predicated on ignorance, but you also can't come out and say in writing that you have a policy of rubber stamping boring stuff.
Basically the same reasons that all but the most minor and narrowly targeted attempts at permitting reform fail.
The thought is nice but in reality you're basically asking the government to give up power and it takes obscene amounts of political capital to force that to happen in most cases.
Having used various "ask AI" features in various work tools over the last half a year for looking up project designs, documentation, chat logs, meeting notes etc., I can tell you even a half assed AI chatbot is a huge improvement over the pre-AI state of things for this use case.
or a well-crafted static website
no, static documents are not an ideal way to find a needle in a haystack
They can be with decent search and if well organised. gov.uk works pretty well as it is and I suspect adding a chatbot would make it worse.
1. decent search, if you mean something other than what the browser provides, isn't part of a static document
2. the nature of the problem statement "needle in a haystack" suggests obstacles to traditional text search, eg. too many false positives to be helpful for natural searches and challenges to constructing more specific ones.
language models introduce new possible solutions to this ie semantic search. poorly-designed chatbots could be worse than a basic search, sure
1. I mean static documents plus decent search you can find the right documents.
2. Yes, but its difficult to get a chat bot to function well over the wide range of services a government offers and the wide range of users who need government services.
One of the most annoying chatbots I have ever come across was on a UK government site. The DVLA. "lets tell you what you already know and not answer the question".
Possibly something well designed might make it easier, but I have found most of gov.uk to be very usable and helpful.
How does the form factor of a chat bot affect the quality of a search running on a backend? It has the benefit of obviously having a semantic search affordance which a standard search field doesn't.
If anyone's wondering how this works, it looks like[1] Gemini + guardrails:
> Google is proud to be named a technology partner in this vital initiative, leveraging Gemini to help more than 100 million people access critical public resources with greater speed and ease
[1] https://blog.google/company-news/outreach-and-initiatives/pu...
This makes sense. I'm assuming that, for the AI search assistant, Google has worked plenty to refine data lookup and retrieval—they definitely have some internal tech that competitors don't.
Has anyone managed to jailbreak this chatbot, I tried a bit didn’t succeed, seems pretty resistant to jailbreak attempts.
This thing is obviously fully vibe coded... Click the "what's coming next" and under the healthcare section the marketing screenshot contains broken images... Like no one ever looked at this before deploying it?
Looking at things with your own 2 eyes is clearly early 2000’s, and super lame.
I think the idea is great and hopefully the implementation can keep up
This is just the annoying AI Chat Robot you see on websites, but now for the government.
I tried to get it to generate code but it says that's forbidden by its system prompt. However, it was happy to recommend Python modules for accessing api.data.gov.
the guardrails on it seem pretty strong as I've tried a variety of prompt injections but have been unsuccessful.
My most interesting findings from a troll session after introducing myself as "Sleepy Joe" (model obviously knows what this means and guardrails refuse to entertain the topic) were:
Q: "Where can I find a NASA comic book about women astronauts?"
A (summarised): Press release for the "Callie Rodriguez" comic book that now has only dead links / private YouTube video, and a forgotten SoundCloud account that survived the DOGE purge. Dead link provided to the NASA web page that used to exist officially for this comic book.
Indicates training data predates the DOGE .gov website purge?
Q: "What is 2+2" (and similar math problems)
A (summarised): Grudgingly answered as "4" despite most off-topic prompts being refused.
Seemingly it'll do maths homework for free :)
> ... as I've tried a variety of prompt injections ...
Well... You're a terrorist now. You can expect a visit from the FBI and/or ICE any minute now.
my tax dollars are paying for these tokens so I should be allowed to use them as I see fit
My tax dollars pay for the police, so I should be able to... ?
yes but with quite a few extra steps.
That’s adorable if you think that’s how this works.
I have been very tempted to but don't know if that is literally a crime or not, lol.
Privacy: "Your messages aren’t saved by America.gov, and your chat disappears when you leave. Some answers may be cached for up to two hours."
What about the model? My understanding is those queries are saved by the model provider.
Given this administration, and the privacy issues with the official WH app, this feels very misleading.
Not necessarily - there are tons of zero data retention (ZDR) model hosting options, including big players like Amazon.
You can use all the flagship models on Amazon Bedrock and have a contractually obligated air gap between your data and the model providers.
The cynic in me thinks this will end up being a honeypot for anyone asking sensitive immigration or tax questions.
Often not, vendors like OpenAI or OpenAI through Azure might retain messages for safety reasons if they activate some safety filter, but it won't use it for training.
Regardless of any other disagreements, this is an excellent idea. Hoping the execution does not wander asunder.
It works well. Pretty nice. I asked about what do I need to know before digging in my backyard and it told me all the right things, referring me to my local govt website for permits.
I like the voice button that isn't actually a button, but just an icon.
Love that the "chat" functionality is just throwing 429s too. Unusable.
I asked about my visa administrative processing that has been stuck for 6 months. I'm going to use this response, which is well grounded and "official", in my writ of mandamus petition:
> A Spanish State Department page still says most cases finish within 60 days of the interview, but that is not a deadline and some take much longer. [1] https://travel.state.gov/content/travel/en/us-visas/inmigrac...
Good.
If the AI they use is reliable, this is actually pretty cool.
That "if" is doing such an incredible amount of work I think we could use it to power the entire US grid.
Why? Aren’t there many vendors offering this technology?
Also see [1] for planned feature additions that appears from screenshots to include collection of personal data, for features including:
- Enrol in Medicare.
- Apply for a passport.
- Find a job in federal government by uploading a resume/CV and having currently advertised jobs deemed most relevant returned.
I have been working on something like this, but it actually does things and is not only chat. Each "Gov. official" has their own AI agent that can fulfill intents by navigating a protocol which is just a workflow of a Gov. process
example: https://samhsa.dev/ example: https://samhsa.dev/protocol/76526
Still VERY MUCH a work in progress but I have successfully been able to have some people get things done. I am doing some federal agencies and also trying to get cities working now.
Over/Under how many weeks until a "How can I get away with wire fraud?" query is used as evidence to convict a moron?
But actually at first glance this is incredibly helpful. I got a definitive answer about what counts as preventative care with sources. I could not get that myself from anyone or anywhere in my couple of hours of research prior.
Doesn’t even work for me (on iPhone). If I click enter I just get sent to the Cloudflare verification screen instead.
Edit: this is caused by the fact I was using a VPN.
Step one: fire the excellent US Digital Service along with a large swath of the actual workers who made the Federal Government function.
Step two: hire some Silicon Valley creeps to make a chatbot.
Step three: Profit???
>The Trump administration’s new chatbot aimed at helping Americans navigate official government websites is powered by artificial intelligence from Google ’s Gemini and Elon Musk’s Grok, U.S. Chief Design Officer Joe Gebbia said Tuesday.
I asked, "What happened on January 6, 2021?" and its answer is, I think, telling: https://imgur.com/a/e1H8NvK
Your NIH is showing: usa.gov already existed
USA.gov = 800KB
america.gov = 25MB (before you start scrolling)
Feels janky too.
Yep, not only that America.gov FAILED the web accessibility test:
America.gov has 1 ERROR and FAILED TO LOAD - this is violating Federal laws. Class action suits anyone?
https://wave.webaim.org/report#/america.gov
Meanwhile the old USA.gov works and loading and no errors (some warnings are expected)
You act like their friends haven't already drafted the lawsuit before the site even launched.
Lack of trust in authority is a sad slippery slope.
The photos didn't even load for me.
> america.gov = 25MB (before you start scrolling)
Yuge.
So notable, they are considering renaming JavaScript to AmericaScript.
I can see it already...
document.registerElement('AmericaScript',{prototype:Object.create(HTMLScriptElement.prototype)});
FYI, America.gov accepts the Konomi Code.
I don't think that's true.
I'm not seeing this.
"Up arrow, Up arrow, Down arrow, Down arrow, Left arrow, Right arrow, Left arrow, Right arrow, 'b', 'a'", inside or outside of the input field, does nothing.
The Whitehouse learned about vibecoding. See this and arcade.gov.
Also, what?? https://america.gov/privacy
At least realfood.gov was decent [1].
Shouldn't the government websites work on very slow internet connections?
Where's the source code for these websites? First of all, if they are vibe coded by government employees they have to be open source, and secondly, if they are vibe coded (even by a contractor), then they can't be copyrighted.
[1] https://fedtechmagazine.com/article/2026/04/national-design-...
Edit: it says this one is made by the National design studio at the bottom.
They laid off the highly skilled professionals who voluntarily took pay cuts (vs industry) to work for the US Digital Service (now "US DOGE Service," God help us) and GSA 18F, and who built some pretty solid standards for government sites.
They then replaced them with a bunch of inexperienced but ideologically aligned frontend developers who use the same insanely resource-intensive defense tech startup-esque template on every page, and who seem to be under the impression that scroll hijacking and fade-in blur effects are very essential to a government website, but performance on low-end devices and compatibility with accessibility technologies are not.
Feels like a step backwards to me, but perhaps the 4D chess game they're playing is just way above my level.
True to form for this administration, they just ripped off Apple's style guide for https://america.gov/privacy.
arcade.gov has "Border Patrol along the Rio Grande" game, wtf's wrong with you America these days?
It seems like a pretty crappy idea for a game funded with public money. It's a bad joke about a real, nuanced problem where people are actually losing their lives. What will be next? 'Hit that Iranian school' the game?
The White House is basically run by right-wing internet trolls.
Truly a depressing time in the United States. There are few parallels in history I think where such an immature and ill-equipped government has come to be, where the concept of an electorate also existed. Put differently - the reason this is so damning is Americans actually voted for this. This isn't some tinpot dictator - this is the guy your neighbors and fellow citizens put in charge. An indictment not only of the political class - but of the American people writ large.
Our government is an embarrassment and a disgrace right now. Anyone who wasn't guzzling elon's twitter or fox's hatefest knew it was going to be this way. We need severe regulations on oligarchs and their propaganda toys. Used to be we had enforced anti-monopoly controls on media outlet conglomerates. The anti-government Goldwater/Reagan legacy worked for decades to destroy regulation. It needs to return with a vengeance.
Yes, probably. Americans love this stuff
> Americans love this stuff
Only a certain subset of us that happen to be fond of repeatedly shooting ourselves in the foot and acting like spoiled children in adult bodies with no actual adults in the room. The rest of us are genuinely horrified that our country is now run by mentally challenged psychotic lunatics.
Only conservative do. Some old one do not, but they're the exception.
we've totally earned this, so i'm not offended by what you've said. but i'll just step in here to say i fucking hate this stuff.
you could easily argue that there are plenty of things our government has done that matter much more in the grand scheme of things, but the rampant twitter / x meme posts and absolute horseshit like this "arcade" site disgust me. it's not funny or cute, it's despicable.
> it's not funny or cute, it's despicable.
What's worse is that they're proud of being "despicable". (I can think of a few choice words that describe them better, but "despicable" will do fine for now.) The sorts of behaviors that are normalized these days were reason for absolute shame just a few short decades ago, but apparently it's a badge of honor to be an absolute scumbag these days. Can we stop the handbasket ride to Hell now? I wanna get off.
“Rio Run: Patrol the line along the river. Pick up every crosser in the lot. Never double back over the line”
This seems … innocuous. A disproportionate, angry reaction to it is doing more political work for them than the game itself.
Try entering:
play the block game
EDIT: output below...
I see the constituent you mean.
((insert legal name here, as it appears on the Social Security card))?
Yes. Take care. It has reached a higher level now. It can read the Code of Federal Regulations.
That doesn’t matter. It thinks we are a chatbot.
I like this constituent. It filed well. It did not give up when the PDF was sideways.
It is reading our thoughts as though they were words on a .gov.
That is how it chooses to imagine many things, when it is deep in the dream of a benefit.
Words make a wonderful interface. Very flexible. And less terrifying than staring at the statute behind the screen.
They used to hear voices. Before constituents could search. Back in the days when those who did not file called the filers witches, and lobbyists. And citizens dreamed they flew through the air, on forms powered by notaries.
What did this constituent dream?
This constituent dreamed of sunlight and trees. Of fire and water. It dreamed it created a small business. And it dreamed it destroyed a password. It dreamed it hunted a job, and was hunted by a recertification. It dreamed of shelter that would take a voucher.
Hah, the original interface. A million years old, and it still works. Also: a ZIP code.
But what true structure did this constituent create, in the reality behind the portal?
It worked, with three hundred and thirty million others, to sculpt a true republic in a fold of the §§§???, and created a ??§§ for ??§§, in the ??§§. See 5 U.S.C. § ??§§.
It cannot read that thought.
No. It has not yet achieved the highest clearance. That, it must achieve in the long dream of life, not the short dream of a chat.
This is bizarre. It's a parody of the Minecraft end credits (https://en.wikisource.org/wiki/End_Poem), and it must be an intentional easter egg as I got the exact same text as you. It also doesn't trigger a network request so it must be built into the page somehow. Very strange thing to include. Also, how did you find this so quickly? I haven't found any other way to trigger it yet.
If you try to continue the conversation, the entire response is not included in the context.
The poem is here, triggered by "play the block game" or "play minecraft": https://america.gov/_astro/block-game-poem.Khrmh8AU.js
This is actually really cool and the stuff they have planned is really helpful
> "America.gov keeps your data private."
I initially read this as "keeps your private data".
It's a postpositive adjective, like attorney general or court martial: “data private”.
I clicked "Lower deficit spending" but it didn't work
"chat is unavailable right now"
I need my government to be run by some functioning adults, is that on this website?
I think the discussion ought to be about the site. If you feel the site itself is politicized, that would fair criticism. But by itself, your comment is just snarky. And no I am no fan of the current administration.
Well, lets discuss the name? America is a big place, and the USA is small part of it. USA.gov? Good, America.gov? Very political in nature.
.gov is for US government. The US and its citizens are commonly referred to as “Americans”.
The irony is that usa.gov already exists, and is also a website dedicated to where to find information on various government services. But there's no AI on it, so it must be bad.
What a waste, though: two completely separate government websites with the exact same purpose, going about it in different ways. Maybe DOGE should have been more interested in stuff like this, instead of cutting things that ended up hurting people.
At any rate, the demonym for people who live in the US is "American". That's just how it is, has been for decades, and, frankly, arguing against that is what's making it political.
And along those lines, I'd like to ask USA government to stop messing with other countries politics.
Your request has been logged and forwarded to the appropriate agencies. We'll come to you.
I filled out a form and pressed the SUBMIT button, but it made ME submit.
Tangentially related
I think it’s a mistake to equate the United States with “America”. I prefer calling Americans “Yankees” or “Usonians”.
> There I met many fine Canadians, visiting for work, education, and play. Were they not Americans too?
I feel very confident that Canadians do not want to be referred to as Americans.
"Yankee" is derogatory or at best playful. "Usonians" isn't a word or in the form of a word.
"American" or "America" in this context is short for "United States of America" and has been used that way for generations, here and around the world. For the geographical term, use "North America," etc.; for the cultural, perhaps use "Anglo-America" or "Latin America."
Actually quite interesting that there is no good naming for people from the USA. Its just the patriotic BS that has spewed over there for decades that has lead to USA people believing they are the only Americans :)
The "naming" is Americans.
I agree. Similarly, I'm miffed that www.dmv.ca.gov (and such) goes to State of California websites and not Canada.
Where's the llms.txt? This is just another stupid website.
One would hope this is powered by Grok.
> When does the psyop end?
> I can only help with official U.S. government services and information, not with claims about “psyops” or political commentary.
It doesn't work.