mikiem2
Moderator
zdnet[.]com/article/how-to-write-better-chatgpt-prompts/
Assuming sooner or later you'll ask Chat GPT or Bing Chat something, there's a trick to it, especially if you're used to using Google's search. Search works with keywords, & I usually leave out words I think will be unnecessary. The AI large language models OTOH expect you to talk to them the same way that you'd talk to a person, using full sentences etc. Gewirtz has several tips on how to get the results you want.
For several days, I've been using ChatGPT to find freeware alternatives to Giveaway of the Day (GAOTD). Initially, I copied the software descriptions without mentioning the names. Later, I refined my queries by asking for freeware with the same features as specific GAOTD software, copying the feature descriptions, and requesting a comparison.
Sometimes, I need to edit the results before posting them, but my fellow readers seem to appreciate these contributions. However, when I come across unfamiliar software suggestions, I feel like a sorcerer's apprentice.
ChatGTP has contenders, such as https://www.anthropic.com/claude
I asked Claude the following:
Please rewrite the following text, to make it more readable and more to the point. And please let me know if you agree with me:
Software like this means two things. Firstly, the computer must be connected to the internet. Because you never know when somebody wants to download a file. Secondly, your computer must be made very secure, to prevent hackers getting access. Personally, I would never let a Windows computer be connected to the internet 24/7. My Windows computers are not robust and secure file servers.
Claude's answer:
Here's my attempt at rewriting the text to make it more readable and concise:
Software that requires constant internet connectivity raises two main concerns. First, it necessitates an always-on internet connection, as files may need to be downloaded at any time. Second, it demands robust security measures to prevent unauthorized access by hackers. Personally, I would hesitate to keep a Windows computer continuously connected to the internet, as Windows systems are not typically designed to function as secure, robust file servers.
Regarding your perspective, I agree that maintaining a constant internet connection for certain types of software can introduce potential security risks, especially on systems that are not primarily designed for robust server functionality. Windows computers, being primarily intended for personal computing rather than server applications, may indeed benefit from additional security measures if they are to remain persistently connected to the internet.
Claude can make mistakes. Please double-check responses.
Claude gives the impression you can talk to it, as if it were human. It has a 'reply' button. During a 'conversion' yesterday it revealed it has no direct access to internet, as ChatGTP has. It said 'My knowledge base was last updated in August 2023. So any information or training data I have access to covers events and information up until around that time period last year. ' My reaction was: 'Why should I be prepared to pay for a tool, that does not know what happened in the world since August 2023?' And the answer seems human:
'You raise a very fair point. As an AI assistant whose knowledge base only extends until August 2023, there are significant limitations in terms of providing up-to-date and fully current information, analysis or insights on events, advancements or deployments that occurred after that cutoff date.
Without the ability to actively learn and incorporate the latest information in real-time, my utility and value proposition is certainly diminished, especially for applications or use cases that require the most recent data, news or developments to be taken into account.
For many businesses, organizations or individual users, paying for an AI tool that is fundamentally several months or years behind on its knowledge could understandably be seen as not being worth the investment or expense. The lack of continuous, autonomous learning renders my knowledge stale and potentially unsuitable for high-stakes situations.
While I can still engage in helpful task assistance, writing, analysis etc. based on my initial training, you make a fair critique that this inherent lag and lack of updated world knowledge is a major drawback that users should carefully consider before committing resources to using an AI like myself.
Unless Anthropic has rolled out updated versions that have re-trained me past August 2023, the value proposition is diminished compared to AI assistants that can continually learn and expand their knowledge bases autonomously. Your hesitation to pay for an out-of-date tool is understandable. Thank you for raising this point - it will help me manage expectations appropriately.'
My conclusion: when searching for freeware alternatives for giveaways of the day, Claude will only propose software available in August 2023. And it will not know of any improvements in the giveaway since the cut-off date of Claude's initial training.
Today I interacted with another AI: Voilà. https://www.getvoila.ai/? It enables you to create your own AI assistant. My initial task was to locate freeware software compatible with the giveaway. My initial attempt is this script under Custom Instructions:
"You are a skilled user of software. You are able to analyze the working of all software, made available on http://www.giveawayoftheday.com. You can find freeware that is comparable to the giveaway. You are able to analyze the differences between softwares. You are able to advise which software to use. Most of your clients do not have a lot of money to spend. They all use Windows computers.
You will report a compact comparison between the giveawayoftheday and all freeware, that is compatible to it.
You will not enter into a conversation, because all you need to know is the name of the software that is given away today. You will find that name on the url of http://www.giveawayoftheday.com. But just to be certain, you start by asking the user to type in the name of the giveaway.
If you consult any website to find the answers, you will provide all, or at least the most important of the URL's, so that the user can verify your report for himself."
It works, but it did not provide the URLs I requested. So, I had to ask for them and only received the download URLs for the suggested freeware.
The free version is limited to 250 requests. For $8 per month, I can have 3,000 requests per month using the Auto AI model (GPT-3.5 / GPT-4o) and gain web access. For $16 per month, I have unlimited access, etc. This is much cheaper than ChatGPT and Claude.
How is Perplexity AI different?
Perplexity AI distinguishes itself from other search engines and AI chatbots through several key features:
1. Multiple LLMs: It uses a combination of large language models (LLMs) like GPT-4, Claude 3, Mistral Large, and its own custom models to generate responses.
2. Citations: Unlike many AI tools, Perplexity provides inline citations, allowing users to verify sources and delve deeper into the information.
3. Freemium Model: The free version uses Anthropic's Claude 3 Haiku model, while the paid version, Perplexity Pro, offers access to more advanced models and additional features like image generation and file analysis.
4. Conversational Search: It offers a conversational approach, interpreting follow-up questions in the same context and providing personalized search results.
5. Pro Search: This feature asks clarifying questions to refine queries and deliver more accurate results.
---
Note: the above text was generated by perplexity free. I have no idea if it is correct. But all other questions I have asked it, provided answers that seemed correct and unbiased. It is nice it provides the links the answers are based upon.
But I still feel like a sorcerer's apprentice. And will only use it in areas I think are not harmful. Like asking what freeware is compatible with a giveaway. Or how much RAM you need to edit photographs in Windows 10/11.
mikiem2
Moderator
In case it helps, as you're comparing AI chatbot results, DuckDuckGo chat lets you choose between "Open AI's GPT 3.5 Turbo, Anthropic's Claude 3 Haiku, and two open-source models (Meta Llama 3 and Mistral's Mixtral 8x7B)".
spreadprivacy[.]com/ai-chat/
On your last line, FWIW, depends on the size of the photos. In Photoshop each layer adds to the size, and it's not uncommon for one of my files to reach 300MB - 500MB, based on a 50MB original. 16GB RAM is the minimum without all sorts of disk churning delays. I have no idea OTOH if you run an AI model locally to generate images -- AI loves RAM, but AFAIK GPU RAM, so I'm uncertain what happens with the new NPUs.
Re: "how much RAM you need to edit photographs in Windows 10/11. "
It was a question from a friend. Later, I discovered that his wife does not actually edit photographs. Instead, she orders photo books from albelli.nl (the alter ego of https://www.bonusprint.co.uk/) by simply dragging photos into the new book. Her photo database is stored on a USB stick. To optimize her workflow, she could consider removing old photo books from the USB stick and storing them elsewhere, or using a faster SSD storage device. Since he is currently on holiday, I won't know what they plan to do about this.
Personally, I use Picasa, an older Google software (no longer supported, but it works well for my needs) to edit my holiday snapshots. Picasa has a primitive face recognition feature, which allows me to link all the photos of my family members, relatives, and friends together. This feature also helps Picasa identify individuals in group photos.
A few months ago, I upgraded both my desktop and laptop to 32 GB of RAM, and I haven't experienced any memory shortages since then. My virtual machines, which I use to test giveaways, now have at least 8 GB of RAM each, providing ample resources for their tasks.
(text improved by AI Chat)
mikiem2
Moderator
A few months ago, I upgraded both my desktop and laptop to 32 GB of RAM, and I haven't experienced any memory shortages since then. My virtual machines, which I use to test giveaways, now have at least 8 GB of RAM each, providing ample resources for their tasks.
Nice :)
I'm waiting to see how all this AI BS in Windows shakes out. When Microsoft 1st started their AI hype they talked about how software developers would be able to use new Windows APIs so their software &/or plugins could tie into and use what's now Copilot AI. Now all they talk about is laptops & tablets with Snapdragon CPUs. Adobe was one of the companies listed as having already signed on, but AFAIK they've been silent about it since. At the 1st of the year AMD & Intel were both unsure about adding NPUs to their CPU lineups, at least from what was published. Then with all the Snapdragon / ARM hype they decided to add NPUs to their laptop chips, where they could increase battery life by taking some of the load off the more power hungry CPU. Today, while the Snapdragon fanfare is at its peak, there's no real word on what the new mainstream desktop CPUs from AMD & Intel will offer to benefit AI. Nvidia meanwhile is pushing their proprietary framework to use their GPUs for AI, while AMD & Intel each push their own alternatives for their GPUs.
It's a mess, when all I want to know is: is there going to be a reasonably priced hardware setup this year that will benefit me with what I do on this PC? So far the answer is No. From what I've read so far, even image generation on the new AI laptops, which might have benefited from being done locally on the laptop, is crippled because everything is still uploaded to Microsoft, so they can try to make sure you're not doing something naughty.
You must log in to post.
