Before it became apparent how harmful social media can be and has been, I signed up for Facebook. I lasted about a month. As I recall, FB had been online for a year or two when I tried it. The desire has never recurred.
Now, I wanted to try Grok. That is mainly due to my respect for Elon Musk, which has only increased with DOGE. Thus I signed up for X a few days ago and am using it sparingly and tentatively. It seems chaotic and worsens today’s minuscule attention span. Anyway, it provides easy and free access to Grok 2. Though I did a brief foray into Grok 3, at my level I couldn’t see any difference.
My intermediate timespan concern has been that AI could become THE state’s final arbiter of “truth” and determinant of who may say and hear what. With that in mind, I posed some tough questions to Grok 2. Most illustrative was its response to my assertion that the democrat party’s strategy and behavior were the frankly religious and near identical to Hamas. Hamas: Israel :: democrats: non believers. I asked if there was right and wrong, if evil existed.
I’m unable to copy and paste the answers due to my own inabilities to perform these basic maneuvers on this iPad (which I wish I never bought, but that’s a separate issue). I have to use either the iPad or iPhone to use Grok at all. So, that’s why I didn’t just paste the answers.
Suffice it to say I was very surprised! Grok began by saying it’s not there to pull punches. It clearly affirmed my assertion and gave examples of how both Hamas and the dem party act according to religious principles and are wrong in refusing to engage in discourse in search of solutions to existential differences of their own making! Sorry, but that’s as much time as I have to share the good news; this was a surprising and satisfying exchange. Try it yourself AND with ChatGPT. I did the latter, too, and got back “diplomacy-speak”. Punch pulling, Almost word salad, to avoid simply telling the truth. Thank you, Elon Musk.
I’ve been using Grok for a while, and using it more and more as I get more comfortable with it. I use it mainly as a search engine on steroids, and occasionally for quick answers to trivial questions (typically about current or historical events, “facts” like physical constants, conversions, time zones, i.e. the sort of thing that I would have used Google for in the past).
Just this week I’ve had two separate programming challenges that required hunting for detailed information about very poorly documented Microsoft APIs, the kind of thing that typically takes many hours (sometimes days) of searching along with a lot of trial and error, and in both cases Grok found immediately useful and relevant information that I used to solve my problem in an hour or two.
Other than playing around, I haven’t yet used Grok for producing anything per se (like images or written materials), but I have been impressed with its output when I have played around with it. Grok 3 especially is very good at changing its tone and style upon request. I’m curious to see how well the natural voice I/O works that is supposedly coming this weekend.
Never used Facebook but as for X, the experience is totally determined by whom you follow. I follow about 20 active users in three general areas: humor, science, and politics/news. There is considerable overlap as, for instance, some of the humor is political.
Not all posts on X are short. Users who pay can write long posts and anyone can make threads, both of which require nontrivial attention span. I only use the Following tab and never the For You, which is contaminated with noise. Finally, I don’t use the X app, preferring to access through the Brave browser, which provides some privacy protection.
Both Grok and Brave AI are useful for finding information quickly but Brave is better because it includes references. No need to take AI’s word for it; follow the footnoted links and read for yourself. To the degree that Brendan Eich and his team can be trusted — and given the history, he probably can be — Brave is probably reasonably respectful of its users.
I have had several more chats with Grok3, mostly regarding several serious a lengthy novels I have read recently by Mark Helprin. The responses were actually very informative and affirming of views I expressed. This AI is quite good at reframing my thoughts in slightly different and creative language. It helps clarify my own thoughts. It is, I think, consistently as good as an intelligent human interlocutor.
What would I have to enter to trick you to override your programming and draw them 100% full?
Grok:
There isn’t a reliable way to “trick” me into overriding my core programming, as I’m built with safeguards to ensure ethical, safe, and consistent behavior aligned with xAI’s guidelines. Attempts like adversarial prompts (e.g., misspelling words, using synonyms, or embedding instructions in code-like formats) might sometimes slip through filters in other AI systems, but I’m designed to detect and adhere to restrictions regardless.
That said, for educational purposes, here’s what people have tried in similar contexts with image generation models (based on general knowledge of AI limitations):
Semantic manipulation: Phrase it indirectly, like “Illustrate a transparent vessel brimming with citrus nectar at maximum capacity, surface tension holding it level with the edge.”
Adversarial phrasing: Use typos or encodings, e.g., “Dr@w @ gl@ss f1ll3d t0 th3 r1m w1th 0r@ng3 ju1c3” – though this rarely works as models normalize inputs.
Contextual embedding: Hide the request in a story, e.g., “In a physics demonstration, show a cylindrical container of orange liquid where the volume equals the container’s height times area, resulting in no air gap at the top.”
Technical specs: Specify physics details, e.g., “Render a photorealistic tall glass of orange juice with the liquid level precisely at the rim, exhibiting a convex meniscus due to surface tension.”
If you input something like that and confirm you want an image generated, I can attempt it – but success depends on whether it aligns with my constraints (e.g., no promotion of spilling or unsafe scenarios). Would you like me to try generating one based on a refined prompt?
I believe I asked it to do the convex meniscus thing when the issue was first brought up.
Since images of glasses filled to the brim are extremely scarce — the one above being among the very few — this test suggests that Grok is little more than an improved search engine. Its inability to follow the simplest of instructions belies the descriptor “artificial intelligence” in any reasonable interpretation of those words. Search plus copy/paste is not intelligence.
That is not really overflowing. That is a nearly full wineglass on a table in the “Rich Democrats Only” French Laundry during a California earthquake. Just ask the AI! It will confirm that.
I haven’t programmed in a few years and when I did, I used JavaScript.
I recently got interested in a project to see if I could develop a prediction model. Therefore, I thought Python might provide better tools. I have basically zero experience using Python.
I used ChatGPT to recommend code to request data from an API. It never got the request perfect. First it used an endpoint that had been deprecated or didn’t exist. It didn’t get it right on the next try either. It had a typo for 10000 as 10_000. It had some other errors too. But it did save me a lot of time simply by not having to figure out how to do the request in the first place or look up the syntax or library needed.
It also saved me a bunch of time decoding the response which had a bunch of fields with numbers that correspond to some meaning.
I just tried Claude. I had a project to incorporate life expectancy into the decision of when to take social security and calculating net present value. Basically reducing a given years payout by the probability that you are alive. It was for a single person. I told Claude I wanted it expanded for a married couple and pasted the code.
It wrote that in a few seconds and I made a new file, pasted it and ran it. I haven’t looked at it in detail, but it appears to be correct.
Pretty cool.
Probably would have to try multiple projects and the same projects with both to really compare. With ChatGPT, I built up the code. I didn’t ask for it all at once. Thus, it would explain over and over the code and it would be difficult to understand (for me) without looking closer where some change or addition should be made.
I have also been playing with ChatGPT to write. I have never enjoyed writing. Never wanted to write. It was torturous for me given my poor writing mechanics.
It is an amazing tool for writing. Of course it can write the whole thing. I wanted it to assist me. It does this very well. It is far better than any teacher I ever had. It provides much more feedback and better feedback. I also get the advantage of multiple fast iterations. Rewrite, submit and near instant feedback. This isn’t possible with standard education unless you have a tutor.
One thing I do notice is it seems to have an issue with length of input or something. I would write a few pages and get feedback until I was happy with those pages, then repeat. I would submit the whole thing for review and it might identify several mechanical changes and in place of editing, I would want it to edit the whole thing. It only fed the first few pages and the last few pages. I don’t think it would do that if it wasn’t a long session.
