Do you have anything written up about how you're doing this? Curious to learn more...
1,259 karma · joined May 6, 2009
Contact: <HN username> + "x" @ gmail
Likewise, I've felt like the meritocracy story that the author sets up as the "moral foundation" has heavily attenuated in this century. It's still used as the justification in America (I'm rich because I deserve it, you're not rich because you didn't work as hard/smart as me) but it feels like that story is wearing thin. Or that the relative proportion of the luck / ovarian lottery aspect has become so much larger than the skill+hard work aspect.
The trend of the rich getting richer, of them using their power to manipulate the system to further advantage them and theirs at the expense of everyone else, existed before AI burst into the public in '20-21. Maybe, like the fake media, it will finally be the kick people need to let go of the meritocracy trap* and realize we need a change.
* I like the notion of meritocracy, it just seems like America has moved from aiming for that, to using the story of it as an excuse to or opiate for the masses.
(There are Hatzolah organizations all over the world, where there are Jewish communities.)
That might still be true where I grew up, in the US, but that's certainly not a guarantee in Melbourne, where I now live. On joining the local volunteer organization, I went from thinking "oh this will be a useful bonus for the community" to "wow, we can literally be essential". Since our org is composed of people living within the community, average response time to ANY call is <5 minutes (lower for cardiac arrest, when people really move). Sometimes one of us is right next door.
We can't do everything an ambulance paramedic can, but we can give aspirin, GTN, oxygen, CPR, and defibrillation. We can also usually navigate/bypass the usual triage system to get the ambulance priority upgraded to Code 1 (highest priority, lights + sirens, etc.) If for some reason the ambulance is far away (it backs up all the time), we can go in the patient's car with them to the hospital, with our gear, in case of further issues in transit.
I tell everyone now to always call us first (since our dispatcher will also call the ambulance) but while I feel more confident in how I'd handle an emergency, I feel less safe overall, with the system's faults and failings more exposed, and the illusion of security stripped away.
My condolences to the author.
In terms of updating - consider whether The System is really working. If not, what can you do yourself (or within your larger network) to better prepare...
So, if anybody else is frustrated and not finding anything online about this, here are a few things I learned, specifically for structured output generation (which is a main use case for batching) - the individual request JSON should resolve to this:
```json { "request": { "contents": [ { "parts": [ { "text": "Give me the main output please" } ] } ], "system_instruction": { "parts": [ { "text": "You are a main output maker." } ] }, "generation_config": { "response_mime_type": "application/json", "response_json_schema": { "type": "object", "properties": { "output1": { "type": "string" }, "output2": { "type": "string" } }, "required": [ "output1", "output2" ] } } }, "metadata": { "key": "my_id" } } ```
To get actual structured output, don't just do `generation_config.response_schema`, you need to include the mime-type, and the key should be `response_json_schema`. Any other combination will either throw opaque errors or won't trigger Structured Output (and will contain the usual LLM intros "I'm happy to do this for you...").
So you upload a .jsonl file with the above JSON, and then you try to submit it for a batch job. If something is wrong with your file, you'll get a "400" and no other info. If something is wrong with the request submission you'll get a 400 with "Invalid JSON payload received. Unknown name \"file_name\" at 'batch.input_config.requests': Cannot find field."
I got the above error endless times when trying their exact sample code: ``` BATCH_INPUT_FILE='files/123456' # File ID curl https://generativelanguage.googleapis.com/v1beta/models/gemi... \ -X POST \ -H "x-goog-api-key: $GEMINI_API_KEY" \ -H "Content-Type:application/json" \ -d "{ 'batch': { 'display_name': 'my-batch-requests', 'input_config': { 'requests': { 'file_name': ${BATCH_INPUT_FILE} } } } }" ```
Finally got the job submission working via the python api (`file_batch_job = client.batches.create()`), but remember, if something is wrong with the file you're submitting, they won't tell you what, or how.
I was referring to 3e, when the "simulation" aspect of the game was more heavily emphasized. See: https://www.dandwiki.com/wiki/SRD:Decanter_of_Endless_Water
> “Geyser” produces a 20-foot-long, 1-foot-wide stream at 30 gallons per round. ... The geyser effect causes considerable back pressure, requiring the holder to make a DC 12 Strength check to avoid being knocked down.
It was that last line that initially sparked the idea. Given the stated effects, this didn't seem like so much of a physics+rules stretch. The no-friction freedom of movement may have been more beyond the pale. Unfortunately 5e deliberately tried to close all the fun ways one could abuse various items.
I actually sat down and worked out all the equations based on the mass of my character and the amount of thrust the decanter provided. Our party would be deep in the wilderness somewhere and I'd say " I nip back to town to pick up some supplies, with acceleration and deacceleration it takes me 17 minutes".
Looking back, I think I was a pretty annoying player, but my DM was very patient. I guess he could see I put a lot of work into the scheme. It was also probably the most exciting application of physics I had encountered in my life so far.
Not all llm errors are hallucinations - if an llm tells me that 3 + 5 is 7, It's just wrong. If it tells me that the source for 3 + 5 being 7 is a seminal paper entitled "On the relative accuracy of summing numbers to a region +-1 from the fourth prime", we would call that a hallucination. In modern parlance " hallucination" has become a term of art to represent a particular class of error that llms are prone to. (Others have argued that "confabulation" would be more accurate, but it hasn't really caught on.)
It's perfectly normal to repurpose terms and anthropomorphizations to represent aspects of the world or systems that we create. You're welcome to try to introduce other terms that don't include any anthropomorphization, but saying it's "just wrong" conveys less information and isn't as useful.
1. Dump the whole textbook into Gemini, along with various syllabi/learning goals.
2. (Carefully) Prompt it to create Anki flashcards to meet each goal.
3. Use Anki (duh).
4. Dump the day's flashcards into a ChatGPT session, turn on voice mode, and ask it to quiz me.
Then I can go about my day answering questions. The best part is that if I don't understand something, or am having a hard time retaining some information, I can immediately ask it to explain - I can start a whole side tangent conversation deepening my understanding of the knowledge unit in the card, and then go right back to quizzing on the next card when I'm ready.
It feels like a learning superpower.
This reminds me of the amazing 2013 video of Travis Rudd coding python by voice: https://youtu.be/8SkdfdXWYaI?si=AwBE_fk6Y88tLcos
The number of times in the last few years I've wanted that level of "verbal hotkeys"... The latencies of many coding llms are still a little bit too low to allow for my ideal level of flow (though admittedly I haven't tried one's hosted on services like groq), but I can clearly envision a time when I'm issuing tight commands to a coder model that's chatting with me and watching my program evolve on screen in real time.
On a somewhat related note to conversational interfaces, the other day I wanted to study some first aid stuff - used Gemini to read the whole textbook and generate Anki flash cards, then copied and pasted the flashcards directly into chat GPT voice mode and had it quiz me. That was probably the most miraculous experience of voice interface I've had in a long time - I could do chores while being constantly quizzed on what I wanted to learn, and anytime I had a question or comment I could just ask it to explain or expound on a term or tangent.
At a time when it seems like so many pursuits or activities or things to make are overshadowed by " but won't there be a model in the next 6 months that can just do this itself?", not to mention all the other present world uncertainties...
Well, it would be nice to hear more thought as to how to focus one's energies.
(I have my own thoughts on this of course, but what I'm really advocating / hoping for is more strong takes on the question.)
DEXA is definitely cheaper, but a good amount of my time spent in MRIs was due to assisting in various research and QA projects. Unless you're made of money, I wouldn't recommend that to anyone who has to pay. I wish they were cheaper...
(EDIT: Nothing to do with medicare or fraudulent billing. Just pushing back on the "for fun" point. I can fall asleep in those things.)
The nice thing about first-class production sqlite support is that even if you do end up with n+1 queries, it's not as big a deal: https://www.sqlite.org/np1queryprob.html
Certainly I wouldn't care about it while prototyping. I can always go back and optimize queries with judicious `.joins()` or `.includes()` if it becomes a bottleneck.
The biggest pain point was the lack of Grade A documentation for the best way to use ActionCable and Turbo – information is spread out between Rails Guides, API docs, and then the Turbo / Stimulus documentation. The actual API docs do a poor job of explaining basic concepts like "streamables", and I kept wondering if I was doing things the "right"/idiomatic way.
Still, as always, ActiveRecord is my biggest draw for Rails, and the new first-class Sqlite integrations are a huge draw for me. I've yet to find an ORM that allows me to be anywhere near as productive.
My other request is probably not in line with your business model. I get the sense that Autotab is always communicating with some server on your end, probably for the various bits of AI functionality. What I was asking for is the ability to export the actions/workflow as, say, a python script (like a Selenium script, or even better, a script which drives your browser) which performs the actions in the Autotab workflow.
I need AI understanding when creating the workflow, or healing in case of an error, but I don't always need it when just executing a prepared script. In those (non AI needed) cases, I don't really want to use up my runtime minutes just because I'm executing a previously generated workflow.
I tried it out on a workflow I've been manually piecing together and it gave me a bunch of "Error encountered, contact support" messages when doing things like clicking on a form input field, or even a button.
The more complex "Instruction" block worked correctly instead (literally things like "click the "Sign In" button), but then I ran out of the 5 minutes of free run time when trying to go through the full flow. I expect this kind of thing will be fixed soon, as it grows.
In terms of ultimate utility, what I really want is something which can export scripts that run entirely locally, but falling back to the more dynamic AI enhanced version when an error is encountered. I would want AutoTab to generate the workflow which I could then run on my own hardware in bulk.
Anyway, great work! This is definitely the best implementation I've seen of that glimpsed future of capable AI web browsing agents.
You run one command that it generates a single executable that can run simultaneously on Mac Linux and windows. Pretty nice for just deploying simple Python scripts.
WhatsApp and messenger groups don't work for this kind of thing because 1) people are often members of many different groups that they would have to constantly notify if they were "okay" during a particular event and 2) many troubles in the world are ongoing, and constantly spamming a message group saying "I'm still okay" doesn't work.
My app just lets people hit a single button to tell any interested friends / family that they are safe. They can do this as many times as they like.
Normally I would be worried about premature optimization, what I've been spending extra time making the tech stack initially very performant. It's working for my family but once I deploy to the world I want it to be solid and stable, or it loses a lot of its value.
Would be great to have a playback speed button as well. (I can't sit through any audio at 1x.)
I think the key with any kind of self-help advice or book is that you have to study it, not just read it. I plan to be working with this book for at least the next six months. I read too many other "inspirational" books that didn't have a lasting impact; the first read is just research to decide whether it's worth devoting time to. Then the real work begins.