Despite its advantages, ChatGPT 5.1 is not without its quirks

ChatGPT 5.1 arrived final week, bringing higher intelligencesooner responses, improved character modes, and extra. I’ve had the possibility to play with it for about 6-8 hours whole to date, and I agree that whereas it is not a large improve by way of processing energy, it is nonetheless a noticeable enchancment in some ways. In fact, that does not imply it is not with out its flaws or obstacles. That is very true for those who use ChatGPT 5.1 as it’s out of the field.

Beneath, I cowl the most important options I got here throughout and a more in-depth take a look at customization that helps if the default setup would not impress you. Earlier than I get into that, I wish to spotlight the adjustments that stood out in a constructive manner as effectively.

Probably the most spectacular adjustments I’ve present in ChatGPT 5.1 to date

One of many first issues I observed was a change in tone and texture. ChatGPT 5 favored brevity and sometimes relied on bullet factors, emojis, and hyphens as a substitute of clear paragraphs and fuller explanations. GPT-5.1 nonetheless has an off-the-cuff TLDR-leaning model, however the phrase alternative and depth of response really feel completely different. The mannequin tends to supply extra substance with out taking away from the casual tone.

To be clear, these responses haven’t been with out some quirks, which I’ll deal with later. The largest factor is that his solutions really feel deeper and higher in step with my character.

One important cause is that it appears to remind me extra absolutely. GPT-5 was inconsistent with previous conversations. GPT-5.1 is considerably higher. It brings up names, previous occasions, private preferences and different particulars with rather more accuracy. It even follows the stylistic guidelines I give it and not using a downside. For instance, I advised him to by no means use dashes when proofreading. GPT-5 usually ignored this request, even when it accepted the rule in its personal pondering tips. GPT-5.1 has met the requirement in all my testing to date. Might it lastly slide? Perhaps, however that is nonetheless progress.

The mannequin can also be extra prepared to make use of what it is aware of about my habits, pursuits, and character with out prompting. In consequence, conversations really feel extra alive. For instance, I talked about the way it has been simpler to cope with behavioral issues with my cat than in previous conditions with comparable cats. GPT-5.1 identified that my expertise with proudly owning mice over the previous two years could play a job.

He famous that rats are usually threatened and that profitable them requires endurance and persistence. He advised that this expertise would possibly have an effect on the way in which I deal with animals generally. The factor is, the final time we talked about my mice was actually months in the past, on a special mannequin, however the identical account. It is very spectacular, I feel.

It might nonetheless method the sycophantic tendencies of older variations, however GPT-5.1 strikes a greater stability. It may be encouraging and complimentary, nevertheless it additionally pushes again when she thinks I am mistaken or oversimplifying one thing. Whereas I nonetheless want it was a bit of extra direct by default, I additionally perceive that not everybody needs open suggestions.

The primary factor is that GPT-5.1 feels extra like an actual conversational accomplice. It handles context higher, follows instructions extra reliably, avoids skipping steps, and responds a lot sooner.

Now for the lower than good. ChatGPT 5.1 usually gave me higher responses total, however some responses nonetheless felt odd and oddly unnatural at instances.

A great instance got here from a role-playing atmosphere I am constructing for my sons. I mentioned to the mannequin: “They determine to settle in America, since there aren’t any folks to fret about.” Later, she replied with: “We select a continent with out folks” is believable”. That quote was by no means mine, so the phrase appeared unusually contrived although the purpose was clear. Whereas the remainder of the reply was excellent by way of info offered, the wording was a bit obscure.

The patch additionally confirmed some quirks. He relied closely on bullets and shortcuts and generally rewrote greater than crucial. That mentioned, I used to be in a position to repair lots of this by in search of no bullets, no drops, and different such orders.

When it comes to total responses, whereas it matched my character higher and was much less demanding, the tone usually felt extra like a web-based pal than an AI accomplice for critical duties. His sentences have been usually shorter, often clumsy, and he was extra prepared to make use of slang or profanity.

Character mirroring explains a few of this, however simply because I swear every so often does not imply I would like it to echo. I desire a car that stays on the bottom. The mannequin additionally used emojis extra usually, rather more than she would have realized from my tone. I do not thoughts this in informal conversations about science or philosophy, however formal matters profit from a extra structured model.

There was additionally fluff. Phrases like “here is the half most individuals by no means perceive” or little bits of reward popped up the place they added nothing significant. Flatter is best managed than on older fashions, nevertheless it’s nonetheless there.

He additionally appears to have a love affair with bullets that’s just like what the earliest fashions have with em dots. After I requested for a quick World Warfare I breakdown, the GPT-5.1 gave me a dense wall of clearly readable bullets. GPT-5 addressed the identical requirement with a clearer mixture of prose and construction, even when 5.1 supplied deeper factual content material.

The excellent news is that the majority of those options will be mitigated by means of character settings. With 5.1, these changes keep extra dependable than earlier than, which makes tuning a lot simpler.

The customization options appear to have extra impression in ChatGPT 5.1

As you’ll have observed, not the entire options I discussed above will matter to everybody. Folks use AI instruments in numerous methods, and chances are you’ll discover fully completely different options than I do. The excellent news is that ChatGPT now makes it a lot simpler to fine-tune your expertise. Default personalities and customized guides have been already accessible, however this technique has obtained a serious improve.

I examined most of the identical requests in ChatGPT 5 and 5.1 utilizing a number of personalities. Previously, these personalities helped improve the expertise, however they did not really feel as developed as they do now. The upgraded variations are hotter and extra expressive, though the change additionally implies that some fashions reply fairly otherwise than earlier than.

For instance, I beloved the Nerdy character. He tended to be thrilling and passionate whereas nonetheless attending to the purpose sooner than the default model. To indicate the distinction, I in contrast how Nerdy responded to “clarify the principle occasions of WW1 to me” on GPT-5 versus GPT-5.1 within the screenshots under.

I’ve to say, the GPT-5.1 model felt nearer to the Quirky character mod in GPT-5. I just like the extra descriptive and fascinating method, nevertheless it’s not all the time what I would like. Switching to Skilled helped deliver issues again to the extra easy Nerdy-style output I loved on the GPT-5, simply with added element. It was additionally a bit of too boring.

Fortunately, customized directions can improve any primary tone even additional. For instance, I added “heat, exploratory, enthusiastic” to Skilled. This stored the readability of the enterprise model, however added a little bit of character, and the next responses have been richer and higher aligned with what I favor.

I’ve additionally discovered that directions like “do not draw” or “decrease bullets” assist steer the mannequin away from among the quirks I discussed earlier. That mentioned, customized personalities and tips can current their very own quirks. For instance, the enterprise persona would generally begin a response by asserting “Beneath is a transparent, easy, business-style clarification…” and if I included “no fluff” within the personalised directions, he would often reply with traces like “here is a no-fluff reply.”

The great factor is that this conduct was much more widespread in ChatGPT 5, however GPT-5.1 would not all the time really feel the necessity to remind you of your customized directions each time you begin a brand new chat. So it is higher, simply nonetheless inconsistent right here.

ChatGPT 5.1 actually feels sooner, friendlier and extra conscious than ever earlier than. Is that this all simply smoke and mirrors? There are lots of choices, nevertheless it’s exhausting to disclaim that the brand new expertise is extra partaking and vibrant than earlier than. With customization, it is also a lot simpler to get the solutions you are in search of with this mannequin.

In fact, she nonetheless hallucinates. He nonetheless makes use of weak phrases on occasion. It is nonetheless extra sycophantic than I might like, even when that may be reverted by means of character customization. For me, I’ve all the time felt that the GPT-5 was higher than it obtained credit score for, though I do know the 5.1 provides it an additional coat of paint and some different enhancements that assist it stand above the outdated mannequin.

For many who weren’t impressed with GPT-5, this may occasionally very effectively be the improve you have been ready for. That mentioned, those that favor older fashions like GPT-4.1 should still discover this mannequin does little to vary their minds.