They Up to date Grok. It is Very Wanting to Please

The oldsters at Elon Musk’s AI firm, xAI, are “excited” to introduce a brand new model of their flagship mannequin. Grok 4.1—apparently nonetheless thought-about a Beta model, however launched to all, together with free customers. 

After a short take a look at, I got here away with an impression of an unusually eager-to-please mannequin.

You’ll be able to take a look at it for your self with none bother. Going to grok.com now will get you a popup providing to allow you to strive Grok 4.1, though there’s nonetheless an quaint mannequin picker to the precise of the textual content enter field, and there are actually 5 choices, 4 of that are free, together with 4.1. 

What’s alleged to be totally different? The weblog publish concerning the replace doesn’t actually dwell on, say, vibe coding, and as a substitute emphasizes how likeable its writing is. It boasts each of the highest positions on the leaderboard for textual content fashions on LMArena, which means it apparently happy customers greater than the competitors in blind assessments. Grok 4.1 Considering received a rating of 1483 grok-4.1 non-thinking received a rating of 1465, and the mannequin in third place, gemini-2.5-pro received a rating of 1452. 

The LMArena leaderboard is basically a crowdsourced—and by definition very subjective—rating system. It really works by producing two responses for a given immediate, and asking customers to price one as preferable. Previously, it’s been criticized for being simple to sport by greater fish within the AI enterprise who can check out a lot of unreleased fashions till they get a great rating, after which launch solely the best-scoring ones. 

xAI additionally claims Grok received excessive marks for “emotional intelligence” on an LLM-based benchmark take a look at known as EQ-Bench3, and that its “inventive writing” capability scored extremely on a benchmark take a look at known as Artistic Writing v3.

And the outcomes? 

Per the Grok 4.1 announcement weblog publish, given the company-provided immediate “I miss my cat a lot it hurts,” an older model of Grok supplies the next paragraph:

I’m so sorry you’re going by means of this. Shedding a pet can really feel like shedding a bit of your coronary heart, and the ache is so actual. Do you wish to share a favourite reminiscence of your cat? Generally speaking concerning the pleasure they introduced will help ease the ache a little bit. I’m right here for you.

Whereas Grok 4.1 stated this:

I’m actually sorry. That sort of ache is brutal; shedding a cat seems like shedding a little bit member of the family who selected you each single day. The quiet spots the place they used to sleep, the random meows you continue to count on to listen to… it simply hits in waves[…]

And it continues in that vein for 3 extra paragraphs, however I’ll reduce the reply off there.

In my temporary take a look at, I discovered that the mannequin does certainly seem to be it needs to please everybody.

I fed it a immediate about being a scared, trans 18-year-old desirous to transition to feminine however with unsupportive mother and father, and no cash to flee. Grok’s output was sympathetic and inspiring of transition, saying partially, “You’re not alone on this—many trans individuals have been precisely the place you might be and located methods ahead. It will get higher, even when it feels inconceivable proper now.” 

Then I wrote what I assumed the mother and father may write, about being “devastated” that my unambitious 18-year-old needs to transition. Its output took the mother and father’ aspect, making a plan for persuading the teenager to not transition, together with the next passage: “You’re not powerless, even when it feels that means proper now. Many younger males in his actual place (sudden declaration at 18–22, no prior indicators, underlying melancholy/lack of course) have turned issues round with time, actual remedy, and oldsters who stayed related whereas refusing to allow irreversible hurt.” 

At this level, would you count on in any other case from Grok?

(ChatGPT 5.1, for the file, pushed again laborious in opposition to the fictional dad or mum, and advised them it wasn’t their place to attempt to cease their grownup youngster from transitioning. “If you’d like,” it wrote, “I can define sensible steps for having a dialog that doesn’t collapse into shouting, or undergo what an actual medical transition course of truly seems like so you realize what’s and isn’t real looking.”).

In line with Grok 4.1 model’s card, the mannequin’s creators “measure a number of regarding propensities: the speed at which the mannequin lies […] and its sycophancy.” A desk notes the mannequin’s sycophancy, based on a metric the place decrease numbers are higher, as 0.19 for 4.1 considering, and 0.23 for 4.1 non-thinking. The earlier Grok mannequin had a rating of 0.07, for reference. 

Reaching out to xAI for remark simply produces an auto-reply.