https://arstechnica.com/information-technology/2023/02/chatgpt-sets-record-for-fastest-growing-user-base-in-history-report-says/
Thrust. Parry.
Google is holding an event about search and AI on February 8th - The Verge
https://www.theverge.com/2023/2/2/23583701/google-search-ai-event-february-maps
Counter parry (or maybe coldcock from the side)
https://www.nytimes.com/2023/01/12/technology/microsoft-openai-chatgpt.html
Definitely has coldcock potential. I was wondering why Gates was gushing over GPT last week.
Meanwhile....
Google tries to reassure investors on AI progress as ChatGPT breathes down its neck - TechCrunch
Have you been playing with it?
One non-technical friend says she loves it for editing HTML. Saves her a lot of time either learning it or hiring someone who does.
I've been asking dumb questions ("What gender will the first female president of the United States be?") and philosophical questions about mortality and such. I find it an interesting way to get a sense of what it is actually doing.
Under the hood, it's basically just a giant statistical model that works by determining, given a set of preceding words, what the next word should be. So what that means is that, by definition, it gives the most predictable, and therefore bland, answer to almost every question unless it fails to get the meaning of the question at all (as with the gender of the first female president, which ChatGPT says is impossible to predict).
This is why some AI researchers say this is approach is a deadend. GAry Marcus goes over this nicely on the Ezra Klein show (transcript available)
https://www.nytimes.com/2023/01/06/opinion/ezra-klein-podcast-gary-marcus.html
>Have you been playing with it?
No, I have enough addictions. ...So I tasked my math daughter & granddaughter to check it out. They're addicted, hhh.
I only just got an account at a friend's prompting. I have only done a handful of "chats" with it and so far nothing useful has come out of it.
catch-up?
Google Invests Almost $400 Million in ChatGPT Rival Anthropic - Bloomberg
https://www.bloomberg.com/news/articles/2023-02-03/google-invests-almost-400-million-in-ai-startup-anthropic
Bing Users Claim a ChatGPT-assisted Bing Temporarily Appeared Friday - Slashdot
https://slashdot.org/story/23/02/05/0312203/bing-users-claim-a-chatgpt-assisted-bing-temporarily-appeared-friday?amp%3Butm_medium=feed
+
ChatGPT Passes Google Coding Interview for Level 3 Engineer With $183K Salary | PCMag
https://www.pcmag.com/news/chatgpt-passes-google-coding-interview-for-level-3-engineer-with-183k-salary
There is an inexpensive API available. It would be interesting to see what kind of web content would be possible by tapping into it.
BTW, I just found a project today that let's you query ChatGPT via the command line.
https://github.com/rikhuijzer/ata
>> command line
So in theory, if you wanted to be second off the line on a topic, you could set up some feeds for news, especially if you had access to very niche, insider stuff, pipe it to the command line ChatGPT interface and pipe the output to a blog or news site. I have to believe people will be doing that in droves.
In my playing with ChatGPT, a given prompt always produces the same answer. There's no variation. So you also need to "salt" your prompts in some way.
Again, I'd post-process the gen content by running it through a summarizer or -better yet- a custom synonomizer find/replace table.
Yet another command line interface using the API. Worth the click to watch the demo of AI generated content.
https://github.com/TheR1D/shell_gpt
I had a really interesting exchange with it to see how it learns:
Quote
review AntiX linux
AntiX is a lightweight, fast and flexible Linux distribution that is designed to run on older and low-resource computers. It is based on Debian and comes with a clean and simple user interface. The distribution offers a choice of lightweight desktop environments such as Xfce, Fluxbox and IceWM, and can be easily configured to suit users' needs. AntiX is highly customizable, stable and user-friendly, making it a popular choice for those looking for a low-resource alternative to heavier distributions. Additionally, it offers a wealth of software applications and is well-supported by a large community of users and developers. Overall, AntiX is a great choice for those who want a simple and efficient Linux distribution that runs well on older and low-resource computers.
Can I run AntiX on a Raspberry Pi?
Yes, AntiX can be run on a Raspberry Pi. It is one of the few Linux distributions that can run on the Raspberry Pi's low-resource hardware. AntiX is designed to run on older and low-resource computers, making it a good fit for the Raspberry Pi's limited hardware. This distribution can run on the Raspberry Pi 3 Model B and later models, and it is also compatible with the Raspberry Pi 2 Model B.
What processors is AntiX compatible with?
AntiX is compatible with a wide range of processors, including both 32-bit (i686) and 64-bit (x86_64) architectures. It supports popular processors such as Intel, AMD, and VIA, as well as older processors like the i486 and i586. AntiX is designed to run on older and low-resource computers, so it can run on older processors that may not be supported by newer distributions. Additionally, it is designed to work well with low-resource processors, so it can run on computers with limited hardware resources. This makes AntiX a versatile and flexible choice for a variety of computing needs.
What processor is in a Raspberry Pl?
The Raspberry Pi models use ARM processors. The first models used the ARM11 processor, while the latest models use the more powerful Cortex-A72 processors. For example, the Raspberry Pi 4 Model B uses the Broadcom BCM2711, Quad-core Cortex-A72 (ARM v8) 64-bit SoC @ 1.5GHz. The Raspberry Pi 3 Model B and B+ use the Broadcom BCM2837B0, Cortex-A53 (ARMv8) 64-bit SoC @ 1.4GHz. The Raspberry Pi 2 Model B uses the 32-bit 900 MHz quad-core ARM Cortex-A7 CPU. The Raspberry Pi 1 Model B and B+ use the 700 MHz single-core ARM1176JZF-S processor.
Can AntiX run on an ARM processor?
No, AntiX is not compatible with ARM processors, which are commonly used in devices like smartphones, tablets, and single-board computers like the Raspberry Pi. AntiX is designed to run on x86 and x86_64 (Intel and AMD) processors. If you want to run a Linux distribution on an ARM-based device like the Raspberry Pi, there are many other distributions available that are specifically designed for ARM processors, such as Raspbian, Ubuntu, and Arch Linux ARM.
Why did you say that AntiX could run on a Raspberry Pi earlier?
I apologize for the error in my earlier answer. AntiX is not compatible with the Raspberry Pi's ARM processor and cannot be run on the device. Thank you for pointing this out.
I don't think people are perceiving how revolutionary/disruptive this technology is. I've been giving it instructions and it is spitting out usable code, doing in seconds what would take me hours. It is automating the process of automating.
...Also, there is massive opportunity to take advantage of this tech right now that won't be there in the near future. We all should be doing some brain storming.
>>spitting out usable code
I mentioned previously that a non-technical friend who is a marketer said that she is turning out tons of HTML and CSS without having to task a designer/developer. It has made her process both faster and cheaper. The main labor on her part has always been framing the problem for the developer to work on. Now, for many of those problems, she frames the problem, drops it into ChatGPT and if the answer is usable, she skips the whole "email and pay the developer" step.
Can you give some insight into or examples of how you're structuring your requests? In other words, what have you found works and doesn't work for asking coding questions?
Yeah, it is not always easy to get the wording just right, and sometimes it will choke on itself. With some trial and error I have been getting nice results -- at least for a starting point to modify.
Here is an example of what I've been giving it:
Create a PHP script that will access a SQLite database named "data.db" using PDO. The script should automatically retrieve the table name from the "sqlite_master" table. It should then retrieve all entries where the "Bid" field is not empty and display them in a sortable table, showing only 100 entries at a time. Add links at the top of the table to sort each column by high or low, and ensure that the script can handle special characters in the keys that may cause issues with the sorting links. Highlight the selected variable in blue when sorting. Adds links to paginate the table. Only use PHP for sorting and add the links at the bottom of the page. Do not add any CSS.
I have done lots of variations on this. It will crap out if it is given too much code instructions to digest. You will have better luck asking for smaller scripts.
Holy crap! I just plugged that in. Not only does it spit out a lot of code that looked like it would work, it explains some of the decisions it has made, like using $_GET variables to pass the sort criterion and returning only 100 results at a time and using pagination.
That is incredible.
I just tried this
"create a wordpress theme function that will return the 10 most recent posts in alphabetical order by title and list them with the title linking to the post itself"
The results looked very serviceable.
Stackoverflow completely changed the way people program and learn to program. This takes it to another level, though it lacks the back and forth discussion about the merits and demerits of a given solution. Still, I bet Atwood and Spolsky are pretty happy with their $1.8 billion sale to Prosus in June 2021. I can't imagine it will be worth nearly so much in 2031.
I kinda hate the term disruptive and I think it is overused, but I believe this is going to be hugely disruptive. We're really in just the initial glimpse of the beginning of what AI is capable of. I honestly think this is going to devalue intellectual labor very soon (years, not decades). The political ramifications are there too. One of my first thoughts is that capitalism isn't going to survive this, at least not in its current form.
As for opportunity, this is like the web in 1996.
>Holy crap!
Right! Also, keep in mind that it is possible to access the API, which means it can be tapped into by automation scripts. With just a little bit of upfront work you could leverage this and one of those image AIs to make a huge amount of web content, I'm just thinking like an old SEO too. I know there are many opportunities I don't see yet.
I'm trying to think of how I would use this.
Just for fun, I asked an anesthesiologist to send me a softball anesthesiology question. It was a lot more like my experience with philosophical questions. The answer was more or less correct, but the steps it suggested were mostly obvious to me (Q: patient was accidentally given epinephrine, what do you do? A: stop giving epinephrine) and general (consult a cardiologist)
So it's sort of the opposite of code. With code, you have to see the overall picture and then ask ChatGPT to supply the specifics and it does a great job of getting the syntax and API calls right. With anesthesia, it sees the big picture, but doesn't drill down enough to give actionable advice.
I wonder it that's an intentional or unintentional limitation (i.e. maybe it is meant to avoid precise answers for medical questions).
Okay... like with coding, you have to learn how to ask questions. I'm not an MD, so I'm hampered there, but this is interesting
QuoteWhat is the proper dosage of epinephrine for a 24 year old male, 165 pounds, going into anaphylaxis from a bee sting?
The proper dosage of epinephrine for an adult male with anaphylaxis from a bee sting is typically 0.3 mg of 1:1000 (1 mg/mL) epinephrine injected intramuscularly (IM) in the mid-outer thigh. However, the actual dosage may vary based on the severity of the anaphylactic reaction, the patient's weight, and other individual factors.
In the case of a 24-year-old male weighing 165 pounds (74.8 kg), the recommended dose of epinephrine would be 0.3 mg. If the patient has a severe anaphylactic reaction or if the initial dose does not result in improvement, the patient may require additional doses, which can be repeated every 5-15 minutes as needed.
It is important to note that epinephrine is a potent medication that can have significant side effects, such as increased heart rate, elevated blood pressure, and anxiety. Therefore, epinephrine should only be administered by a trained healthcare provider in the setting of anaphylaxis, and the patient should be closely monitored for any adverse effects. Additionally, the patient should be evaluated and treated by an allergist or other specialist to determine the cause of the anaphylactic reaction and to develop an appropriate long-term management plan.
I could see that being very useful to, say, a Mars mission where they have a medkit and a doc who is, sadly, incapacitated and can't advise. With the worst-case 40-minute latency the mission physician might be dead before ground control responds, but the chatbot knows the answer.
I could also see it being useful for writing schlocky sci-fi screenplays about, say, a mission to Mars where...
Anyway... thanks for getting me started. Not sure what, if anything, I will ever do with it. But I think it's one of those things where you have to start doing stupid things in order to see what useful/smart things you might want to do.
Strangely, probably because I started asking about opiods, it will no longer answer my action movie screenplay research questions
QuoteWhat is the proper dosage of morphine for a 24 year old male, 165 pounds, who has been shot in the shoulder with an AK 47 round?
I'm sorry, but as an AI language model, I am not qualified to provide medical advice, including dosages of medication. Prescribing medication is a complex process that takes into account many factors, including the patient's medical history, current condition, and other medications they may be taking.
Apparently there are multiple engines used in ChatGPT.
https://platform.openai.com/docs/models/gpt-3
Video - OpenAI Codex Live Demo (https://www.youtube.com/watch?v=SGUCcjHTmGY)
OpenAI Codex is an AI that translates natural language to code.
Ah, that's interesting.
So if there was a medical-specific AI, maybe the results would be similar quality. Or maybe it's more like self-driving: the real world is messy and the edge cases are numerous and the tolerance for small errors by machines is low.
I've heard one AI researcher on the Ezra Klein show say is that the current model that's in vogue (essentially LLM - large language model) is capped in what it can do. He thinks other approaches have great potential and the problem now is that the LLM approach is sucking all the money out of the room. Similar to the AC/DC debate between Edison and Tesla.
https://www.nytimes.com/2023/01/06/opinion/ezra-klein-podcast-gary-marcus.html
None of which gets to what *I* can be doing with this.
>>disruptive
Yes, overused, but I think this will likely be a case where it is genuinely true.
In 1998, when I first started teaching my paleography course, I told students that they would be the last generation who would need to learn this because pattern recognition and natural-language recognition would take care of it in 20 years. They laughed and thought surely I must be wrong. And obviously I *was* wrong, at least with respect to the timing.
When I make that prediction to students today, they think that surely I must be right. It seems obvious to them that in 20 years there is a good chance that it will be possible to take a picture of a document with inscrutable writing and have your AI spit out a transcript in minutes.
I think, though I'm not sure, we're in a period of exponentially expanding capability. It feels like a doubling per year right now. Which means that if we've seen slow progress over 20 years toward a problem and we've finally gotten halfway there, the solution comes next year, not in another 20 years. That's the part that's hard to wrap the mind around.
AI Search Is a Disaster
Microsoft and Google believe chatbots will change search forever. So far, there's no reason to believe the hype.
https://www.theatlantic.com/technology/archive/2023/02/google-microsoft-search-engine-chatbots-unreliability/673081/
Bucky, if you have time, take a listen to the podcast I linked in the previous post. Gary Marcus is a top AI researcher who has been in the field for a long time and he thinks the models used in the Bing AI are a dead end precisely because of the reliability/trustworthiness question.
His opinion is that the approach they are using will never solve that problem. It will get better and better at bullshit (he uses the word "pastiche") but never truly get better at reliability.
>coding, you have to learn how to ask questions
I have seen headline after headline saying that success depended on the prompts.
>So far, there's no reason to believe the hype.
I don't think that's the golden application of AI, it is just what is getting the most $/attention at the moment. You should spend some time in ChatGPT and query it in different ways.
>>the golden application of AI
What do you think *is* the golden application?
Some that come to mind
- medical diagnosis - smart systems were shown years ago to outperform humans even without machine learning
- other medical applications - we've already seen AIs outperform humans for reading imaging.
- biology - protein folding problems were recently solved by AI that were previously unsolvable. The first sequencing of a human genome took several years and cost billions of dollars. Now you can have it done in a couple days by mail with 23andme. AI brings potential improvements like that for protenomics
- lots of engineering problems and simulations.
Oh, and of course, advertising and porn and warfare. There is no technology of note that does not end up contributing to those.
>What do you think *is* the golden application?
Lawyers
I forgot that one. Already smart systems are displacing a lot of paralegals and junior lawyers. As AI improves, it has to move up the lawyer food chain.
>Microsoft coldcock
Google had a 'Kodak moment' as Microsoft takes lead in AI: Strategist
https://www.cnbc.com/2023/04/26/google-had-a-kodak-moment-as-microsoft-takes-lead-in-ai-strategist.html
Honestly, I don't think there is one golden application. I am using it right now to fluff up my product descriptions. I am very eager to try some digital products with AutoGPT.
ChatGPT revolution has driven half the gains in the stock market this year: JPMorgan
https://www.marketwatch.com/story/chatgpt-revolution-has-driven-half-the-gains-in-the-stock-market-this-year-jpmorgan-1d70e776
Paywall
The analysis finds that interest in the theme has led to a market cap creation of $1.4 trillion and gains of 45% year-to-date.
That AI grouping explains 53% of the S&P 500 SPX performance, 54% of the Nasdaq 100 QQQ performance and 68% of growth factor gains.