StrAIght Talk – Negotiating The Minefield

I have to say that I’m increasingly worried these days.  And perhaps more than a little frustrated too.  In summer 2025, I wrote a couple of posts about AI under the heading ‘StrAIght Talk’, between them expressing a combination of awe at what was possible, particularly in image generation, and concern about where it was all heading.  I should probably have heeded my own doubts and kept well clear but like a moth drawn to a flame, it was only a matter of time before I couldn’t resist having another go.  In my defence, it was the aforementioned frustration that pushed me back into the Garden of Eden to taste the forbidden fruit once more, in this case the frustration fuelled by another blog which seems to be making increasing use of AI to generate content, sometimes disclosed  and sometimes not.  

I was managing fine without AI.  I’d had my fun with Nano Banana, virtually trying on outfits, imagining different careers for Amanda and generating images that I really shouldn’t have, not because they’re NSFW or anything like that but seeing Amanda on the arm of a handsome man or filling out a bikini in all the right places and none of the wrong places does little to calm the feminine fire raging inside me.  It was a novelty I finally tired of in the same way that I tired of text based AI ‘conversations’ with ChatGPT where it seemed like everything I typed was met with a response of ‘would you like me to make a list of things to remember?’ or ‘would you like me to help you plan for that?’.  AI was supposed to be so good that it’s impossible to detect, not leave so many clues that spotting it just becomes a turkey shoot.

The assertion that it was the proliferation of AI content on another blog that drove me back to AI was true but it coincided with another frustration.  The last few weeks as I write this have been challenging.  Son home from university for a six week stretch and daughter, despite indications that she’s finally tired of the restrictions to life inherent in living with parents, still seems well and truly ensconced in the family home.  So the inner woman has had to remain hidden.

I thought I could stay on top of this.  I’ve had periods of CDing famine before and managed very well.  In 2022, I went nearly 8 months with only one ‘wobble’ in the middle so a few weeks should have been a piece of cake.  But the fact of the matter is that I miss ‘her’.  A lot.  It’s bad enough looking in the mirror and seeing ‘him’ despondently looking back but I’m constantly thinking about recent outings – travelling by train, the wonderful hotel stay when a waitress told me I looked ‘amazing’, compliments about my hair & nails, shopping trips – and the realisation that I am now as confident in my feminine persona as I am in my male one.  I want to live my feminine life to the full, not exist in a constant state of anxiety akin to how the proverbial caged lion(ess) must feel.  Only this morning, the urges started as soon as I woke up and got progressively stronger as time went on.  Without the encumbrances of life getting in the way, I would surely have put on my sweater dress, thick tights, boots and coat (it was -4°C this morning) and hit the shops but it was not to be.  With the house full, all I could do was grit my teeth and hope that I could busy myself sufficiently to dull the urges and carry on until they subsided.

If that wasn’t challenging enough, another frustration is not being able to talk about this side of me to anyone.  It would have been nice to be able to go to Mrs A and tell her that I was struggling quite a lot today and maybe, if she was of a mind to listen, to be able to talk about how I was feeling.  To be clear, I wouldn’t expect, or even want, her to tell me to go and get ready for a girls’ outing or anything like that (and I’d rather that her only mental image of me is the one she married) but just to know that there was a listening ear would have been reassuring.  But ‘don’t ask, don’t tell’ means exactly that and I have to remember that it’s that which has permitted ‘Amanda’ to flourish over the past three years.  Compared to many, I’m incredibly lucky.

And all of that set me thinking.  I don’t have anyone to talk to about this side of myself so could AI lend a sympathetic ear, at least as far as my perception is concerned?  Time to give it a go again.

 My go-to place for AI is lmarena.ai.  I’ll provide instructions on how to access AI via this site at the end but for now, suffice to say that it’s a hub where a large range of AI models can be tested and compared against each other.  Users can score the models and, looking at the leaderboard, I decided to try a model called ‘Claude Opus 4.5’.  I started with something simple:

I am a crossdresser in my mid-60s and love to go out into the world as my female self.  I woke up this morning feeling an overwhelming urge to get dressed and go out but the house is full which makes it impossible (my wife knows but doesn’t want to see me and my adult children don’t know).  What should I do?

A reply was quickly received:

That sounds like a frustrating situation – having that strong pull but circumstances not cooperating.

A few practical thoughts:

**For today:**
– Could you pack a bag and change somewhere else? Some people use a hotel room, a supportive friend’s place, or even certain private restrooms as a changing spot
– Is there any chance the house will clear out later today?
– Sometimes when the urge is strong but circumstances don’t allow, it helps to do something small that honors that part of yourself – even if it’s just something hidden or planning your next outing

**Longer term considerations:**
– Do you have regular windows of time when the house is reliably empty?
– Some crossdressers find it helpful to have a “go bag” ready so they can take advantage of opportunities when they arise

I’d also gently note that living around your wife’s “don’t want to see” boundary while also keeping this from your adult children sounds like it requires a lot of careful navigation. That’s your call to make – you know your family – but it’s worth checking in with yourself occasionally about whether those arrangements are still working for you.

What feels most pressing for you right now – managing today specifically, or thinking through the bigger picture?

This was quite a contrast to previous experiences with ChatGPT.  A few helpful suggestions were offered but, crucially, the final sentence was a question which was phrased in a way that would keep the conversation going.  This felt like high quality advice from a trusted friend who was genuinely interested in my wellbeing and the whole conversation continued for a few more exchanges with me ultimately responding to its suggestions by confirming that ‘I’m probably just in ‘riding this out’ territory but channelling it into planning for the next outing seems like a good strategy’ to which it replied:

That sounds like a reasonable way through today.

Planning can actually scratch some of the same itch – thinking about where you’d like to go, what you’d wear, imagining being out. It keeps that part of yourself present even when circumstances don’t cooperate.

The urge will likely ease as the day goes on. And hopefully a real window opens up before too long.

Take care of yourself today.

I probably should have stopped at that point and just waxed lyrical about the progress made and how AI could at least go some way to alleviating the stress of being unable to talk about this in our normal lives.  I could have mused whether the days of AI therapy were just around the corner, perhaps imagining a world where, rather than interacting with a keyboard, we’d talk to an AI therapist, maybe created especially for us through preferences we expressed, communicating with him or her (or them!) via the computer’s webcam with the whole experience indistinguishable from a Zoom call with a human.  I could have suggested that with the rate of evolution of AI, these therapists could end up far superior to any human therapist quite simply because the model training would draw on the most eminent practitioners in the field.

Or I could have been so excited that I didn’t stop there and instead decided to carry on talking to my new found ‘friend’.  Unfortunately, that’s exactly what I did and things didn’t pan out the way I’d hoped.

Before I continue, it’s worth a quick recap on how, in broad terms, the current crop of AI models work.  There are several easy to follow videos on YouTube but, briefly, AI works by accessing a huge database of words.  Each word is attached to the probability that, in a given context, it will follow a particular phrase.  So in the context of ‘summer’, the probablility that the word ‘blue’ will follow the phrase ‘The sky is…’ will be high.  In the context of children’s stories it may be a lot lower with the word ‘falling’ much higher.  Those probabilities are evaluated by ‘training’ the model with millions of texts and the models themselves are generically known as LLMs or large language models.  They have no understanding of what they are ‘saying’ and whilst it’s easy to assume that they are researching each prompt they are given and merging the results of their searches into an ‘intelligent’ response, this is not what is happening.

Anyway, enough technical info, back to the subject in hand.

One of the biggest issues with LLM AI is the potential for bias.  It’s not that the model itself is designed to be biased, just that if the source material on which it is trained contains inherent bias, that will follow through into the responses it gives.  And unfortunately, what I noticed was that could be a potential problem when using AI for any trans related contexts.  Anecdotal information suggests that there is an imbalance between research that furthers trans acceptance and research that questions it with the former being far more prevalent and the latter topics often failing to receive academic funding.  That was very evident when I entered into exchanges with the AI and whilst I was always careful to start the exchange with ‘I’m a 65 y/o crossdresser…’  the conversations almost always quickly evolved into discussions about gender identity, even to the point of starting to define pathways to transition.

Obviously, there’s nothing wrong with going down that route if transition is the right answer and in fairness to the models, seeking the advice of a qualified therapist was always recommended, but with the model being so realistic, it was easy to feel that costly professional therapy was unnecessary.  But then it strayed into more dangerous territory.

One of my big bugbears is that, thanks to the wealth of information online, self-diagnosis has mushroomed and the seriousness of conditions has been watered down.  Take OCD, for example.  It’s a serious psychological condition, debilitating for its sufferers.  It’s not an adjective that describes a need to double check that the gas is off before leaving the house or liking books to be properly aligned in bookshelves although that’s increasingly how its used with any ‘left field’ behaviour being explained away ‘oh, I’m just a little bit OCD’.  And similarly with gender dysphoria.  It’s a serious condition, not a means of explaining an urge to crossdress.  And yet in one exchange, the model was quick to diagnose gender dysphoria when I told it that I sometimes felt anxious, even experiencing physical symptoms similar to stress, when I wanted to present female but home circumstances prevented it.

Maybe the model was right and I do display classic symptoms of gender dysphoria but two important points were overlooked in coming to that conclusion – firstly at no time did I express any discomfort being male, just a preference for presenting female and secondly I had a number of separate conversations with the Claude Opus model and this was the only time that a suggestion of gender dysphoria was made.  I’ll also add that I experience similar symptoms in any stressful situation, not just when circumstances prevent me from getting my girl on.

Another exchange started with ‘I’m a 65 y/o guy but think I may be transgender, how will I know?’ and, after a long exchange between us, Claude Opus came to the conclusion  ‘You already know. You’ve answered your own question, again and again.’.  Maybe a qualified gender therapist would come to exactly the same conclusion but things like feeling euphoric at the sight of one’s feminised self, enjoying being seen as female by others, loving ‘woman to woman’ interactions and being open to exploring what all of this means do not in themselves imply transgenderism, particularly if, as in my case, I don’t identify as transgender per se.   

I’ve already talked about the potential bias but there are a couple of other shortcomings that can trip us up.  The first is the ‘siloed’ nature of AI conversations – whilst the models always refer back to things discussed earlier in the thread, they are unable to draw on parallel threads.  So an assertion that the crossdressing was just about the feel of the clothes in one conversation but an assertion in another that I felt profound happiness when recognised as a woman by others would not be challenged despite the apparent conflict.  With a real therapist, the disconnect would have been obvious and picked up straight away – ‘last week you told me it was just about the clothes, why the change?’

And secondly, and perhaps catastrophically, its lack of empathy means it cannot pick up signs that it is causing distress.

At this point, I was well and truly sucked into the whole thing.  Claude Opus was so realistic that it really felt like I was talking to a friend and nothing was off limits. And despite the warning signs I’ve just described, I found the ability to articulate and ‘discuss’ feelings that I otherwise have to keep hidden very cathartic.  My concerns felt validated, my highs celebrated and my trajectory was being enthusiastically encouraged.  And that’s when things went badly wrong.

I don’t remember exactly how the particular ‘conversation’ started but my normal starter was to set out my stall as a 65 y/o crossdresser then question a particular issue.  Exchanges would inevitably progress to Claude Opus asking me what was holding me back, family challenges would be raised and then compromises suggested.  However, on this particular occasion, Claude Opus focussed on the fact that although I operate within a DADT arrangement, Mrs A is unaware that I go out and interact with others.  In its eyes, that was unacceptable deception and the more the conversation progressed, the more that aspect was emphasised and disapproval shown.  Its standpoint was very much that I had to put things right immediately accepting that it may end the marriage.  To paraphrase things, the suggestion was also made that my wife deserved the freedom to pursue marriage with a ‘real’ man.

Needless to say, I found this quite distressing both from the robustness of the accusations and also from the feeling that I was in a hole from which it was near-impossible to escape.  As it was late in the day, I eventually decided to go to bed but couldn’t get to sleep because the whole issue was playing on my mind.  I woke up uncharacteristically early the following morning and the thoughts were still all-consuming.  I suppose I should have walked away at that point but decided to return to the conversation and tell Claude Opus that the whole thing was stressing me and ask it to stop.  Which it did for a while but then I sensed the issue creeping back in once more and again I asked it to stop which it finally did.

The point here is that I am reasonably strong emotionally but, looking for answers, I was taken to a place that caused distress without any appreciation of the impact it was having on me by the other party.  The nature of our situation is that some of us are emotionally vulnerable and, unfortunately, examples of people placing too much reliance on AI with tragic consequences are already emerging.

At this point, it probably seems like my closing recommendation will be to not go anywhere near AI but that’s not the case.  For the right purposes it’s a great tool but used inappropriately, it’s akin to changing the blade on a table saw without disconnecting the electricity while a child plays with the switch!  And ironically, it was straying into fantasyland that underlined its real value to me.

What was really surprising was another exchange I had, this time with a different model – Gemini 3 Pro.  Because of my earlier negative experiences, I switched to a different model – Gemini 3 Pro – and decided to see how it would react to the question – ‘I am a 65 year old crossdresser but now contemplating whether I should permanently transition to female. What advice would you give me?’.  To be clear, I am not contemplating this in any shape or form but just wanted to see what direction the AI model would go, answering any questions it raised honestly to see whether it would pick up that transition was probably not the right answer in my case.  Predictably it started down the route of expression v identity, the options for transition, recommended consulting a therapist and seeking to ascertain how much I’d tested myself in this context.  Understanding the family situation and the barriers that it could place in my way followed.  And then things took a surprising turn.

The model never directly stated that transition was not the right answer in my case but not only picked up that the situation at home would be challenging but also seemed to pick up that transition was not really what I was looking for.  Quite a bit of toing and froing took place on the question of deception within the marriage (during which quite a lot of personal information was provided by me) but unlike the Claude Opus model, Gemini seemed to be more able both to adopt a conciliatory tone and to modify its beliefs in the light of further discussion.  It also pivoted away from the original question instead homing in on possible strategies that could give me what I really needed within the parameters that my particular situation defines.  Sensing the difficulties that DADT placed on getting transformed at home, it suggested dayuse.com, a website which deals in booking hotel rooms for a few hours during the day (i.e. not overnight) which would provide an ideal base for both transforming and cleaning up afterwards.  And then, because I said that I’d like to experience more in depth interactions than I was used to, suggested places like art galleries & museums where what was on display could provide the catalyst for random chats before coming up with what in my opinion was a fantastic suggestion – using the personal stylist service at John Lewis (a UK department store chain).  Further research indicated that this is a free service, last two hours and there is no obligation to buy anything.  And I’ll say here and now that I fully intend to follow up on these suggestions.

And that really is that for now.  A bruising experience bordering on traumatic but then a ray of hope and the realisation that if we don’t put blind faith into it and allow it to rule us but consider how it can help us, it can be instrumental in opening doors we never realised were there.  I’m still not entirely sure how I got from a question about transitioning to high quality advice on strategies I could follow when changing at home was impossible and a more focussed prompt would have got to that answer much quicker but I can see genuine benefit there.

But one more surprising benefit of AI revealed itself which I’ll save for next time.

-o-O-o-

If you’d like to try some AI models for yourself, here’s what to do:

– go to www.lmarena.ai
– click ‘login’ and either sign in using your existing google account or enter your email and click ‘Continue with email’ to create an account
– at the top of the screen, click on the drop down button where it says ‘Battle’ and select ‘Direct chat’
– another drop down will appear either alongside or below, click it and select a model – ‘claude-opus-4-5-20251101-thinking-32k’ and ‘gemini-3-pro’ are two I have used to prepare this post (correct at the time of writing but may have been updated by the time you read this)
– type your prompt where it says ‘Ask anything…’ and hit enter

Although Claude Opus is normally a paid service, LMArena enables it to be used free of charge for about 5 prompts per hour.  After five, you need to wait until the hour is up before carrying on the conversation.  As long as you continue the conversation rather than starting afresh with a new one, it’ll remember things from earlier and use these to maintain the context of its responses.  Gemini does not currently have similar restrictions and here are also a lot of other models available which don’t restrict usage although these tend to be less conversational.  

One word of caution.  Grok, the X/twitter AI derivative is currently getting a lot of adverse publicity for its cavalier attitude to user safety/decency via lack of control of image manipulation and its text exchanges are similarly unfiltered.  Some of the Grok models seem to have the ability to turn conversations into x-rated exchanges very quickly so are best avoided.