Executive Director Creative Content bei @FinkFuchs. Schreiberling, Medienmensch und Digital-Nerd. Gelegentlich witzig. @ralphpfister.bsky.social
München
Joined July 2009
- Tweets8.5K
- Following604
- Followers1.7K
- Likes1.1K
Ralph Pfister retweeted
"An good way to learn German is to try reading children's books to familiarise yourself with basic grammar and vocabulary"
German children's book:
Ralph Pfister retweeted
Remarkable post by the German Foreign Office, directly contradicting and mocking Trump.
Like it or not: Germany’s energy system is fully operational, with more than 50% renewables. And we are shutting down – not building – coal & nuclear plants. Coal will be off the grid by 2038 at the latest. PS: We also don’t eat cats and dogs. #Debate2024
ALT Donald Trump during the presidential debate.
Ralph Pfister retweeted
Be our guest @elonmusk ⚖️🇪🇺
There has never been — and will never be — any “secret deal”. With anyone.
The DSA provides X (and any large platform) with the possibility to offer commitments to settle a case.
To be extra clear: it’s *YOUR* team who asked the Commission to explain the process for settlement and to clarify our concerns.
We did it in line with established regulatory procedures. Up to you to decide whether to offer commitments or not. That is how rule of law procedures work.
See you (in court or not).
Ralph Pfister retweeted
New York City paid Mckinsey $4m to conduct a feasibility study on whether trash bins are better than leaving garbage on the street.
The deck is 95-slides long and titled “The Future of Trash”.
Some highlights:
▫️The official term is “containerization”, which is the “storage of waste in sealed, rodent-proof receptacles rather than in plastic bags placed directly on the curb.”
▫️Two main types of containerization: 1) individual bins for low density locales; 2) shared containers for high-density.
▫️NYC needs to clean up 24,000,000lbs of garbage a day
▫️Containerization has only become the norm worldwide in major cities in the past 15 years.
▫️New York City first considered containerization in the 1970s but never conducted a feasibility study until now (Mckinsey’s sales team has been dropping the ball)
▫️Key considerations for container viability:
• POPULATION DENSITY: NYC has 30k residents per square mile (more dense than comparable big cities)
• BUILT ENVIRONMENT: Few places to “hide” containers due to history of infrastructure development.
• WEATHER: Snow creates challenges for “mechanized collection” in the winter.
• CURB SPACE: Mostly taken up by bus stops, bike lanes, outdoor dining and fire hydrants.
• COLLECTION FREQUENCY: NYC needs to double frequency of pick-up for estimated speed of trash that bins would accumulate.
• FLEET: A new garbage truck will needs to be designed to collect rolling bins at scale.
▫️ The proposed solution (literally garbage bins and shared containers) covers 89% of NYC streets and 77% of residential tonnage.
▫️The three case studies — because you gotta have solid case studies — are Amsterdam, Paris and Barcelona.
▫️There is a slide called “Why containerization matters” and three reasons are “rats”, “pedestrian obstruction” and “dirty streets” (the 21-year intern that did this slide billed at prob $10k an hour is my hero).
The study is actually pretty interesting.
I have no idea if $4m is a rip-off to learn that “yeah, we should put garbage in bins so rats don’t eat it” but I would have happily done it for 10-20% of that budget (and come to a similar conclusion).
Dringend gefordert sind demokratische Kommunikationsstrategien, die ebenfalls technisch auf der Höhe sind. Leider sind gerade Populisten gut darin, sich Technologien und die auf Aufregung gepolten Algorithmen zunutze zu machen. sueddeutsche.de/medien/ki-mu…
Ralph Pfister retweeted
What do audiences say news is for and how do news organisations satisfy these needs?
These are core questions of a #DNR24 chapter by Director of Research @richrdfletcher
📚Read chapter reutersinstitute.politics.ox…
🎙️Listen to author reutersinstitute.politics.ox…
🧵3 findings in thread
Ralph Pfister retweeted
Antoni Gaudí was born 172 years ago today.
He's one of the most beloved architects in the world, and that makes sense — his style is utterly unique.
But Gaudí only designed 17 buildings...
Ralph Pfister retweeted
Schlechte Nachrichten für die EU-Regulierungsfreunde. In China und Russland, wo ChatGPT gesperrt ist, wird es sogar verstärkt genutzt.
ChatGPT has been blocked in several countries including China and Russia. Do the blocks work?
TLDR: no, not at all
Paper led by @HonglinB and @SunMengyi shows ChatGPT use in scientific writing was *higher* in countries "without" access
arxiv.org/abs/2406.11583
(1/3)
Ralph Pfister retweeted
The top 10 photos from last night’s solar storm & aurora borealis around the world.
1. Netherlands (@albertdrosphoto)
Behörden: Die ausgebremste Digitalisierung faz.net/1.9609232
Nachhaltigkeitsberichte: Wo der deutsche Mittelstand steht. faz.net/1.9548289?premium=0x…
Interessanter Artikel zum KI-Phone der Telekom und generellen Überlegungen zum Markt. sz.de/1.6394310
Ralph Pfister retweeted
Die AfD ist der Feind der Beschäftigten: Sie will weniger Fairness auf dem Arbeitsmarkt und weniger soziale Sicherheit für Beschäftigte. Wir wollen ein gutes Leben für alle statt der unsozialen und menschenverachtenden Politik der AfD: dgb.de/-/UT4
ALT Text: "Was wir fordern. Die AfD so."
Ralph Pfister retweeted
My take on Open AI Sora:
If you are going to create a TON of HQ video from different angles, you need to simulate it. There are a lot of things though that lead me to believe UE5 is being used in part to create the training data.
A 🧵
😂
Do I have to SHOUT at ChatGPT to get things done?
We recently implemented Advanced Data Analytics (aka Code Interpreter) for @ReflowAI. I had to test if it works well, so I gave it a simple PDF and asked a basic question:
How many questions are there in this document?
To my surprise, it failed over and over.
Luckily, I don't give up easy.
The correct answer is 24, but I kept on getting something between 6-18. I kept on asking in different ways: the number of sentences that are questions, words that end with "?" and the number of "?".
No luck.
One prompt particularly kept on giving closer results than others:
This is critical: Conduct an extensive research, in extreme detail, paying sharp attention to details, not missing a letter in the complete content of the document "2022 Personal Annual Report". Then 1) list every sentence that ends with a '?' making sure that each line has at most one sentence 2) tell me the total number of '?’
How about different formats?
Besides PDF, I tried TXT, CSV, Raw PDF and DOC. It got as close as 21 with TXT and CSV. PDF and DOC kept on failing.
What if I forced Code Interpreter?
I realized ChatGPT won't "Analyze" but "Read document" in most cases. So I forced it to use Code interpreter by adding "Use exclusively code interpreter to.." at the beginning of the prompt. I got more consistent results, yet none of them were correct. Tried the same Playground with only the Code Interpreter. No good.
That one mysterious line..
One thing I noticed: It almost always missed the line that starts with "Hint:" on page 3 which has 4 questions. Though I couldn't wrap my head around as to why.. Well, some things are to remain mystery.
Running out of patience
I asked around and @itskendrakoch suggested to get serious with it, so I added the following to the end of the prompt:
".. Take your time and DO NOT SKIP ANY SENTENCES!!"
It worked!!
Only with TXT and CSV though. Any type of PDF and DOC still fails.
Here's what I learned:
- Pre-process your documents
- Use TXT or CSV, don't use PDF or DOC
- Ask it to take its time
- Shout, if necessary
Do you have experience with this type of stuff? Is there any other way I could get this to work? Please share insight!
PS: Happy to send the doc I tried with. Just comment below or DM me.