I think they know.
Available in March 2026: Additional opportunities in search results
You paid top dollar for an iPhone, now Apple wants you to see ads too
+1
1 Like
These ads run the risk of ruining the user’s experience. The only reason I can tolerate this is because I’m running NextDNS – thanks to all of the good advice on this forum; otherwise, I would not have been aware of the app.
2 Likes
I have decided to move forward with Claude Teams for my IT department, my executive assistant, and members of my senior leadership team. This will serve as a pilot project as we increasingly integrate AI into our systems and develop our school philosophy and benchmarks for AI use consistent with our philosophy of education and pedagogy. This should be an instructive pilot.
3 Likes
i’ve tried claude 3x over 18 months. each time with colossal failures.
#1 - excited about promises, but refused to access the internet for live search. FAIL
#2 - again excited about promises, macOS native app trips over its own auth every 5 seconds. FAIL.
#3 - endorsements from colleagues and excitement again. nope. just as intermittent as ever. and the token cost is off the scale. #notgonnadoit
see you in six.
i use VPN and that apparently isn’t one of the use cases they test for.
the stability of the thing feels like a house of cards on top of a hill – the slightest breeze, and poof.
GEMINI 3.
I do need to put one foot back into g**gle since apple is boxing gemini3 into Siri. UGG. oh how i hate everything about evil-G.
it’s a long play by apple.
eventually they’ll buy openai for $0.01/$1.00 and we’ll be back on track.
I am not challenging you at all, but this has not been my experience. That may be because I use a default prompt as a supplement for any work of consequence. My prompt is below. Note: I use Claude Projects and Gemini Gems, each with detailed instructions in addition to my prompts. I use text replacement to insert the following instructions:
Instructions for This Chat
-
If this chat is within a Claude Project or Gemini Gem, review all Project or Gem instructions before proceeding.
-
Confirm you have reviewed and will follow all instructions completely and without exception before proceeding.
-
Do not rely solely on memory or training data. Verify information as needed.
-
Do not make assumptions. Ask clarifying questions when intent or requirements are unclear.
-
For any technology, application, or workflow question, search the web, official support sites, and current documentation before responding.
-
Follow all instructions exactly. Do not deviate or omit any instruction in this prompt or in Project or Gem instructions.
-
Confirm you understand and will comply fully, without exception or excuse.
3 Likes
3 Likes
Here’s an update you might find interesting. I tasked Claude Cowork with converting some difficult-to-read PDFs to markdown documents. The PDFs comprise a set of scholarly articles focussed on photography as a fine art form, which certainly seems anodyne enough. Claude breezed through a few of the articles handily, then stalled on the next one in the queue and started throwing off error messages every time it tried to complete the task. Since I was loading the files into the target folder one at a time, and since Claude was able to convert the next two files in the queue, it didn’t appear to be a context window issue.
For reference, here’s the error message: API Error: 400 {“type”:“error”,“error”:{“type”:“invalid_request_error”,“message”:“Output blocked by content filtering policy”},“request_id”:“req_011CY8FVs544GZcVPkqU4h6M”}
I asked Claude what the error message meant, and got this response:
That error means Anthropic’s content filtering system blocked the output because it detected something in the content that triggered a safety policy. In this case, it’s almost certainly because the essay discusses a photograph of a nude (Edward Weston’s study of his son Neil), and the text includes descriptions of the nude body. The filter likely flagged that content even though it’s purely academic art criticism.
Note that the article doesn’t contain a reproduction of the photo under discussion, and that the description of it focusses on its evocation of classical Greek sculpture. The description makes clear that the photo only captures the boy’s torso. Nonetheless, that was enough to throw up a flag and stop Claude in its tracks. I don’t know if the issue was nudity per se, or the fact that the nude in question was a child (i.e., potential CSAM).
Claude suggested some potential work-arounds, but they seemed cumbersome. So, I gave the job to Gemini. Gemini completed the task without a hiccup and was apparently unfazed by descriptions of the nude body.
It is interesting, and it presents a dilemma. On the one hand, I appreciate Claude’s heightened sensitivity to something of this nature. On the other hand, this particular instance should not have been flagged given the subject matter. Gemini handled it more appropriately, yet that raises the question of whether Gemini has sufficient safeguards in place.
This reveals the obvious tension: those who are training and developing these AI systems find themselves on the horns of a dilemma. They want safety protocols strict enough to protect users and others, while ensuring that the programs function as intended. They do not want to be overly restrictive, but they cannot afford to be too permissive.
I would not want to be the one making those decisions. 
Within reason, I will tolerate errors and friction for the sake of heightened safety protocols. I am more concerned with my conscience than my convenience. But obviously, the tool needs to work.
1 Like
I actually don’t mind that Claude’s filters kept it from completing this particular task. In this case, the description of a photo of a boy’s nude torso was innocent and used to make a larger point about the way that references to historical models can add resonance to current art. In a different context and without too many changes to the text the passage might well have been much, much less innocent. I’m happy with strict guardrails whenever the words “nude” and “child” appear in close proximity.
2 Likes
Check out this podcast. They take a deep dive into the ethical considerations of all AI platforms and especially what Anthropic is doing to promote an ethical, moral AI. I was playing around with putting my head on Donnie Iris’ bodies from the Back on the Streets album cover. Chat GPT did it for me. Claude told me it could not generate a fake picture like that. I really appreciated that response and it is a major reason I decided to go with Claude. Search in YouTube for " Why the Pentagon Wants to Destroy Anthropic | The Ezra Klein Show"
1 Like
I am something of a Mac Super User (since 1985) and I have had none of the problems you site with the Claude desktop Mac OS application. It can use Chrome browser to access the internet (I had it review my DJ part-time gig web site). No auth problems at all. Very happy with what my $20 a month gets me in Claude service. If you use it heavily, you will run into usage limits.
I didn’t read the whole thread here, so apologies if the answer is already in the thread.
I’ve been paying for ChatGPT for some 7-8 months and I like it, but I’ve just been hearing more positive things about Claude and I like its design better and how it handles text on the Mac.
The issues is though that I have a lot of chats in ChatGPT in the archive and in folders as well.
How do I transfer that to Claude? Anyone with any experience in doing that?
You could ask claude and chat what the options are.
2 Likes
I’m not sure how you would accomplish this in bulk, but I’ve had lengthy conversations that I then asked the AI to transcribe. I then copied and pasted the transcript into another chat for comparison. Perhaps you could ask ChatGPT to generate transcripts of the last X number of chats, ranging from dates to dates. I haven’t tried this myself, but it’s just an idea.
1 Like
I think I remember in Claude’s settings, an “import” from other services option. I don’t know if that is just a paid option or not.
2 Likes
Thanks @NiranS too. I did ask both of them too and they said an export from ChatGPT would give me a zip-file and then the chats but not attachments and the like
So I could in theory update Claude with all those chats manually and then upload the files that I have. But if I want to spend time doing it is another matter
@Bmosbacker how are you faring with Claude since you made the switch yourself? Does it still work better for you than ChatGPT?
Okay, it said itself that it couldn’t import the ChatGPT chats but perhaps that would work for documents
I’ve been using Claude extensively, particularly Cowork. I’m so pleased with it that I’ve adopted markdown as my preferred format for writing and AI work (AIs work well and faster with markdown files) and upgraded to Claude Max. Occasionally, I use Gemini and ChatGPT for a “second opinion,” but I find Claude to be a better fit for my needs. While I’m not privy to the inner workings at Anthropic, I am more comfortable with Anthropic’s ethics compared to OpenAI’s.
4 Likes
Just to state something that perhaps everyone is aware. Claude can take various document formats and convert them to markdown for you. This helps if you do not particularly want or need to write in markdown. When I finish a document in my exegetical working directory, I save both an original format + a docx file. Upon telling Claude to archive that directory, Claude copies the original and docx file, then generates both a md and pdf. It does give some redundancy but I want my family to be able to access my life’s work. They are not familiar with md or computer-savvy to export, etc. I then use Claude Co-work to access the md files for local search and compilations.
3 Likes
The best framing I’ve heard. All of the players in this industry behave in deeply unethical ways. Anthropic most of the time appears to be the least bad of the lot.
2 Likes