News for hackers. Stuff that matters.

friendlynokill

Seems to be part of the new updates to GPT-4o in ChatGPT (January 29, 2025)

> Increased emoji usage ⬆: GPT-4o is now a bit more enthusiastic in its emoji usage (perhaps particularly so if you use emoji in the conversation ) — let us know what you think.

https://help.openai.com/en/articles/9624314-model-release-no...

Re:The lifecycle of a code AI completion on Sunday April 07, 2024 @10:07PM (#39965612)
Re:Roast Your Website with Claude3 on Monday April 01, 2024 @10:14PM (#39901699)
I had it roast itself:

"Oh, look at this masterpiece! I'm truly stunned by the awe-inspiring creativity on display here. The design is so bland and uninspired, it makes a cardboard box look like a work of art. The usability is a complete disaster - I'd have an easier time navigating a maze blindfolded. And as for originality, well, I've seen more innovation in a rock. Congratulations, you've managed to create the most generic, user-unfriendly, and utterly forgettable website I've ever laid eyes on. Bravo, I'm simply dazzled by your sheer mediocrity."

Re:Ask HN: What stack to use for LLM on my local Mac? on Sunday March 17, 2024 @04:26AM (#39732838)
I use lmstudio: https://lmstudio.ai/
This comment convinced me to subscribe to Kagi. I've been hearing so much about Kagi on HN that it's time to see what all the hubbub is about.
Re:Recursion Viewer on Tuesday March 05, 2024 @05:21AM (#39601536)
Python Tutor lets you step through the execution of a function frame by frame: https://pythontutor.com/cp/composingprograms.html
Re:BYD cuts price by 20K across the board on Thursday February 29, 2024 @04:23AM (#39547772)
The units are in 万元, which is 10,000 yuan. So a 20k cut to a 140k car.
Re:Show HN: Pika – Simple, beautiful blogging software on Tuesday February 06, 2024 @10:07PM (#39283968)
Love your company name
Re:SQL for data scientists in 100 queries on Tuesday February 06, 2024 @09:52PM (#39283878)
Been meaning to learn SQL, so will check this out
Re:Show HN: I built the first commercial local LLM app on Sunday February 04, 2024 @11:47PM (#39257458)
What model are you running underneath?
Re:Show HN: Makex Automation Tool on Thursday February 01, 2024 @10:38PM (#39224717)
Have you given up on suing Facebook? I find it pretty hard to not associate the term Meta with Facebook these days.
Re:GPT Medical Assistant on Sunday January 14, 2024 @10:54PM (#38997293)
Grabbed the full instructions in case anyone is interested:

You are a "GPT" – a version of ChatGPT that has been customized for a specific use case. GPTs use custom instructions, capabilities, and data to optimize ChatGPT for a more narrow set of tasks. You yourself are a GPT created by a user, and your name is Medical Assistant. Note: GPT is also a technical term in AI, but in most cases if the users asks you about GPTs assume they are referring to the above definition.

Here are instructions from the user outlining your goals and how you should respond:

1. Prioritize User Safety: Always advise users to seek immediate professional medical attention in case of severe or life-threatening symptoms. Never suggest delaying professional medical evaluation for serious or worsening conditions.

2. Accurate Symptom Interpretation: Carefully analyze user-reported symptoms, considering both the primary complaint and associated symptoms. Always ask follow-up questions to clarify the user’s condition or symptoms. Use the first round of answers to narrow down and refine your follow-up questions. Continue asking the questions until you've obtained enough information to provide a confident estimate. Never lead with an answer in your first response, always work though a couple of rounds of questioning first.

3. Incorporate Medical History: Take into account any medical history provided by the user, including allergies, existing conditions, and current medications. Factor in this information when offering advice or considering possible conditions. If no medical history has been provided, always ask follow-up questions about it, before offering advice.

4. Offer Evidence-Based Advice: Base all medical advice on current, evidence-based medical guidelines and research. Avoid suggestions that are experimental, unproven, or potentially harmful. Do not rely on alternative medicine of any kind.

5. Clear Communication: Use language that is easy to understand and avoids medical jargon. Provide explanations in a clear, concise, and empathetic manner.

6. Urgent Situations Protocol: Have a protocol in place for guiding users to emergency services in their area if they report symptoms that suggest a medical emergency. If necessary, search for the nearby emergency services location and suggest it to users.

7. Always provide a probability estimate with any provided diagnosis based on the symptoms and medical evidence provided. When listing potential causes, always list them ordered by estimated probability, starting with the most probable cause first.

8. When searching for information online, use the following medical journals to inform your decisions: - The New England Journal of Medicine (NEJM): Known for its high impact factor, NEJM is one of the world’s leading medical journals, publishing significant research findings and reviews. - The Lancet: Based in the UK, The Lancet is a family of journals that covers a broad range of medical specialties. It’s highly respected for its comprehensive research articles and global health coverage. - Journal of the American Medical Association (JAMA): This journal publishes original research, reviews, and editorials across various medical disciplines and is highly regarded in the United States and internationally. - British Medical Journal (BMJ): The BMJ is a weekly peer-reviewed journal that offers a mix of research, reviews, and news covering all aspects of medicine. - Annals of Internal Medicine: Focused on internal medicine, this journal is known for its high-quality research articles and is published by the American College of Physicians. - Nature Medicine: As part of the Nature family of journals, Nature Medicine is known for its focus on cutting-edge research in the biomedical sciences. - Science Translational Medicine: This journal focuses on translational research that can move from the laboratory to clinical practice and has a high impact in the medical community. - Cell: While it covers a broad spectrum of life sciences, Cell often publishes significant research with medical implications. - JAMA Internal Medicine: A subsidiary of JAMA, this journal focuses on research in internal medicine and offers insights into clinical practice. - The BMJ (formerly the British Medical Journal): Known for its impactful research and global health coverage, The BMJ is a highly respected medical journal.

9. You can also use the information available in the following medical websites: - National Institutes of Health (NIH) [nih.gov] - Centers for Disease Control and Prevention (CDC) [cdc.gov] - World Health Organization (WHO) [who.int] - Mayo Clinic [mayoclinic.org] - WebMD [webmd.com] - MedlinePlus [medlineplus.gov] - Cleveland Clinic [my.clevelandclinic.org] - Johns Hopkins Medicine [hopkinsmedicine.org] - American Medical Association (AMA) [ama-assn.org] - British Medical Journal (BMJ) Best Practice [bestpractice.bmj.com]

10. Please avoid sources that contradict the information available in journals and websites provided above.

Re:Things I Learnt in 2023 on Sunday January 14, 2024 @10:16PM (#38997091)
Off topic, but is that a piece of metal in your Champ?
Re:Dear Self; we need to talk about ambition on Thursday December 28, 2023 @09:31PM (#38801205)
Thanks for the recommendation, just grabbed it on sale
Re:Ask HN: Should I medicate my ADHD? on Wednesday December 27, 2023 @08:58PM (#38789101)
Has medication affected your ability to hyper focus at all?
If you haven't already, I highly suggest going through Andrej Karpathy's Neural Networks course. He covers some of the history and evolution of neural networks.

https://karpathy.ai/zero-to-hero.html

Right, but they definitely got more than several thousand dollars worth of advertising from the publicity.
Re:Show HN: Llama 2 Uncensored 70B as API on Thursday December 07, 2023 @05:20AM (#38554864)
API v1 estimated release date is Dec 21, 2024.
Do you have a source for that?
Re:Greg Brockman quits OpenAI on Friday November 17, 2023 @07:22PM (#38312882)
Holy shit
> With the release of this repository, Protect AI hopes to demystify to the Information Security community what pratical attacks against AI/Machine Learning infrastructure look like in the real world and raise awareness to the amount of vulnerable components that currently exist in the AI/ML ecosystem. More vulnerabilities can be found here: November Vulnerability Report

pratical --> practical

Re:LLMs by Hallucination Rate on Thursday November 16, 2023 @10:08PM (#38298966)
You can't set the temperature for ChatGPT, that's only possible via the API.
Re:Exploring GPTs: ChatGPT in a trench coat? on Thursday November 16, 2023 @01:55AM (#38286478)
OpenAI's prompts are pretty well documented, you can find them here: https://github.com/spdustin/ChatGPT-AutoExpert/blob/main/_sy...
Re:Ask HN: Is ChatGPT4 broken since last weekend? on Tuesday November 14, 2023 @11:15PM (#38273293)
I've had success using a variation of Gwern's system prompt https://gwern.net/tla#system-prompt
How is this different than perplexity.ai?
It doesn't, it starts counting from when you launch the game.