About QumisAI:
QumisAI is the only lawyer-built AI platform transforming insurance knowledge work. Our mission is to provide the most accurate AI solutions for the insurance industry. We are already trusted by five of the 15 largest U.S. insurance brokers, leading specialty carriers, and top law firms.
What We’re Building:
QumisAI leverages cutting-edge technology to tackle complex challenges in insurance knowledge work. From fine-tuned LLMs to proprietary data pipelines, we are shaping the future of AI in the insurance domain.
Why Join Us?
If you’re passionate about:
-Building at the Frontier: Work with LLMs, agents, fine-tuning, classifiers, data engineering/pipelines, and copilots.
-A High-Impact Role: Join a small, ambitious team with the autonomy to make a meaningful impact in a high-growth environment.
-A Career-Defining Opportunity: Shape the future of an industry while defining your own career path.
This is your chance to be part of a fast-paced startup with the stability of being backed by some of the most trusted names in the insurance and legal industries.
Compensation:
Base Salary: $125k–$200k
Equity: 0.50%–1.50%
Contact Us:
If this resonates with you, let’s talk! We are looking for people with "Superpowers"—what are you uniquely the best at or aspire to be the best at? Email us your superpower, and let’s chat: shiv@
(No recruiters. Emails from recruiters will be automatically flagged as spam.)
Superpower: Converts any youtube link into a micro-learning course, extracts the slides, creates bite size concepts, quizzes, AI based answer evaluations.
At Metal http://getmetal.io/ we're currently building a fine tuning platform. We host, index and version embeddings. Provide an easy way to manage the fine tuning jobs as well.
There is nothing more useful than learning what to forget. At least I manage myself for rapidly reshaping my mind to fit the tasks of the year. Despite my postings here, the gcc options for gcc 2.72 don't help much compared to my curre t muscle memory for terraform.
the problem I have with services like Blinkist (or any book summarizing offering) is that you get what the summarizers think is the "point(s)" of the material
You can read, for example, a Christian Apologetics work and come away with tools for public speaking and rhetoric that would be completely missed if you only got the "summary" of the apologetic argument in the religious context
" BI tool so people that don't know SQL can explore data, create/look at dashboards and export data." - We had exactly the same requirement. Use Looker, its expensive but worth it. The requests to our data team has reduced by 90% We are a small startup with 20 employees
It is completely non-intuitive UX which is causing a lot of people issues. Merging branches is the bread and butter of GIT, if you are changing a fundamental primitive (mergers) of how a technology is expected to work, by reversing the direction of the merge request (opposite to the direction provided explicitly by the user) without warning that is a major issue.
It's not really messing with anything fundamental about Git — in fact it's providing a very common workflow for Git users, just in a way that is confusing for people who mainly use Github.
Obviously if lots of people are getting confused something needs to be fixed, but I think this is fundamentally a minor UI wart.
> just in a way that is confusing for people who mainly use Github.
As a user I am requesting to merge branch X into branch Y. What Github actually does is merge branch Y into branch W, with insufficient warning to the user. Gitlab users continue to have issues due to this unexpected behavior. I am astounded this problem has not been addressed in a meaningful way, so users dont counter this issue.
88 versions of the ad, and the first sentence of the ad contains 14 words. The numbers 14/88 have considerable significance here, especially when paired with the use of a concentration camp symbol..
Sure, but with those tattoos, the 14 & 88 are significant and inescapable. There’s no ambiguity and their presence is not hidden. But once you’re counting the number of words in a tweet, you’re talking about things much more hidden. It doesn’t work well as a dog whistle if the metaphorical dogs can’t hear it, either.
Plenty of ambiguity there. I think you may need to evaluate what your motivation is to need to keep disbelieving that racist dogwhistles can actually be ambiguous, especially considering they are that way by design.
Well, along the lines of counting the number of words in a tweet, maybe 14 stars? And then chains, where the linked chains look kinda like 8s? Something where the 14 and 88 are present, but not obvious unless you think to look for them.
That is an mechanism which insulates even well-documented conspiracies that use numerical symbols as identification banners, which is probably why certain of them do exactly that.
The 14/88 miners have so many degrees of freedom in how they search for these numbers they are always guaranteed to find them. If the number of banned ads was < 88 we would be hearing about how some subset of similar looking ads had size 88.
Hardly. It's 14 words, not 14 anything else. Furthemore, neo-Nazis have in fact also used these numbers in their esoteric numerology in varying degrees of freedom.
It's 14 words, in the same way that "And if you check out some of her other tweets on this, such as" is 14 words. And where is the proof that there were 88 versions of this ad? The Media Matters article has a link "ran 88 ads" that points to [1], but there are only 3 ads on that page.
This "evidence" is a Rorschach test. People see what they're predisposed to see.
If you look at the picture of the 3 ads, each has a number in the top left corner showing how many times it ran, apparently generated by the Dewey Square Adwatch tool set cited in the article. The numbers are 30, 28 and 30, which sum to 88. This could have been better explained in the MM article. It could also be a coincidence, but repeated coincidences eventually qualify as a pattern.
So the ad had run 88 times when Media Matters took the screenshot?
Then if it had continued to run, wouldn't it have run more than 88 times? Or possibly it did run again before Facebook took it down, and actually did run more than 88 times?
Now I'm picturing Media Matters watching the ad counter like teenage boys watching an odometer approach 80085.
You say 'of course not' but then just carry on as if the previous poster had endorsed your theory. I am not an adtech person but I had the distinct impression that the tool cited counted the number of promotional spots purchased for each ad placed. On what basis do you say they were 'clearly counting something' as if it were a dynamic rather than a static process?
You're the one alleging that they took the screenshot as it counted up in a dynamic process. I'm asking why you seem so sure about about that, in contrast to alternative possibilities. Seems like you can't answer.
So the ad had run 88 times when Media Matters took the screenshot? [...] The article doesn't specify what is being counted, but Media Matters was clearly counting something. And like any counter, it's bound to pass 88 on the way to however many versions/targetings/updates/campaigns/whatever the campaign had planned to run.
I was willing to listen to your argument, but instead of making a case you're just insulting my intelligence.
Because evidence for something doesn't require proof beyond the possibility of conjecturing alternative explanations, and if it did, nothing would be evidence of anything, since any sense data could just be hallucination, and ultimately anything that could be cited as evidence at some point relies on sense data.
Funny that the first sentence of your comment also contains 14 words. I'm not saying it's also a coincidence in the ad, but trying to do numerology without statistics is pointless. A big pile of examples isn't statistics.
So a big part of right-wing troll culture centers on dogwhistles: something that regular people would see and think, "I don't see what the problem is," but has a less wholesome meaning underneath. For instance, the "OK" hand gesture stuff.
We just got a major textbook example of this with the Trump campaign rally in Tulsa, now scheduled for this Saturday.
Originally it was scheduled for tomorrow, Friday, Juneteenth, in Tulsa, site of the notorious 1921 Black Wall Street Massacre.
Simply vile.
And if the excuse is it was all an honest mistake, even worse. That's the sort of cock-up a grade-school competent campaign team would scrupulously avoid.
Tech Stack: TypeScript | Rails | Postgres | Docker | React | LLM
About QumisAI: QumisAI is the only lawyer-built AI platform transforming insurance knowledge work. Our mission is to provide the most accurate AI solutions for the insurance industry. We are already trusted by five of the 15 largest U.S. insurance brokers, leading specialty carriers, and top law firms.
What We’re Building: QumisAI leverages cutting-edge technology to tackle complex challenges in insurance knowledge work. From fine-tuned LLMs to proprietary data pipelines, we are shaping the future of AI in the insurance domain.
Why Join Us? If you’re passionate about:
-Building at the Frontier: Work with LLMs, agents, fine-tuning, classifiers, data engineering/pipelines, and copilots.
-A High-Impact Role: Join a small, ambitious team with the autonomy to make a meaningful impact in a high-growth environment.
-A Career-Defining Opportunity: Shape the future of an industry while defining your own career path.
This is your chance to be part of a fast-paced startup with the stability of being backed by some of the most trusted names in the insurance and legal industries.
Compensation:
Base Salary: $125k–$200k Equity: 0.50%–1.50% Contact Us: If this resonates with you, let’s talk! We are looking for people with "Superpowers"—what are you uniquely the best at or aspire to be the best at? Email us your superpower, and let’s chat: shiv@
(No recruiters. Emails from recruiters will be automatically flagged as spam.)