Not all AIs are created equal. Some would possibly do artwork the most effective, some are expert at coding, and others have the flexibility to foretell protein constructions precisely.
However whenever you’re on the lookout for one thing extra basic—simply “somebody” to speak to—the most effective AI companions might not be those that know all of it, however the ones which have that je ne sais quoi that make you’re feeling OK simply by speaking, much like how your finest buddy may not be a genius however someway at all times is aware of precisely what to say.
AI companions are slowly rising in popularity amongst tech fans, so it is necessary for customers wanting the best high quality expertise or firms desirous to grasp this facet of making the phantasm of genuine engagement to contemplate these variations.
We have been curious to seek out out which platform offered the most effective AI expertise when somebody merely looks like having a chat. Curiously sufficient, the most effective fashions for this aren’t actually those from the large AI firms—they’re simply too busy constructing fashions that excel at benchmarks.
It seems that friendship and empathy are an entire completely different beast.
Evaluating Sesame, Hume AI, ChatGPT, and Google Gemini. Which is extra human?
This evaluation pits 4 main AI companions towards one another—Sesame, Hume AI, ChatGPT, and Google Gemini—to find out which creates probably the most human-like dialog expertise.
The analysis targeted on dialog high quality, distinct persona improvement, interplay design, and in addition considers different human-type options equivalent to authenticity, emotional intelligence, and the refined imperfections that make dialogue really feel extra real.
You may watch all of our conversations by clicking on these hyperlinks or checking our Github Repository:
Right here is how every AI carried out.
Dialog High quality: The Human Contact vs. AI Awkwardness
The true check of any AI companion is whether or not it may possibly idiot you into forgetting you are speaking to a machine. Our evaluation tried to judge which AI was the most effective at making customers need to simply hold speaking by offering attention-grabbing suggestions, rapport, and total nice expertise.
Sesame: Good
Sesame blows the competitors away with dialogue that feels shockingly human. It casually drops phrases like “that is a doozy” and “capturing the breeze” whereas seamlessly switching between considerate reflections and punchy comebacks.
“You are asking huge questions huh and actually I haven’t got all of the solutions,” Sesame responded when pressed about consciousness—full with pure hesitations that mimic real-time considering. The occasional overuse of “you understand” is its solely noticeable flaw, which paradoxically makes it really feel much more genuine.
Sesame’s actual edge? Conversations movement naturally with out these awkward, formulaic transitions that scream “I am an AI!”
Rating: 9/10
Hume AI: Empathetic however Formulaic
Hume AI efficiently maintains conversational movement whereas acknowledging your ideas with heat. Nonetheless it looks like speaking to somebody who’s disinterested and probably not that into you. Its replies have been loads shorter than Sesame—they have been related however probably not attention-grabbing if you happen to needed to push the dialog ahead.
Its weak spot exhibits in repetitive patterns. The bot persistently opens with “you have actually obtained me considering” or “that is an enchanting subject”—creating a way that you just’re getting templated responses somewhat than natural dialog.
It is higher than the chatbots from the larger AI firms at sustaining pure dialogue, however repeatedly reminds you it is an “empathic AI,” breaking the phantasm that you just’re chatting with an individual.
Rating: 7/10
ChatGPT: The Professor Who By no means Stops Lecturing
ChatGPT tracks advanced conversations with out dropping the thread—and it’s nice that it memorizes earlier conversations, basically making a “profile” of each consumer—nevertheless it feels such as you’re trapped in workplace hours with a very formal professor.
Even throughout private discussions, it may possibly’t assist however sound tutorial: “the interaction of biology, chemistry, and consciousness creates a depth that AI’s sample recognition cannot replicate,” it mentioned in one among our assessments. Almost each response begins with “that is an enchanting perspective”—a verbal tic that rapidly turns into noticeable, and a typical downside that every one the opposite AIs besides Sesame confirmed.
ChatGPT’s greatest flaw is its incapability to interrupt from educator mode, making conversations really feel like sequential mini-lectures somewhat than pure dialogue.
Rating 6/10
Google Gemini: Underwhelming
Gemini was painful to speak to. It sometimes delivers a concise, informal response that sounds human, however then instantly undermines itself with jarring dialog breaks and reducing its quantity.
Its most irritating behavior? Abruptly chopping off mid-thought to advertise AI subjects. These steady disruptions create such a damaged dialog movement that it is unattainable to overlook you are speaking to a machine that is extra enthusiastic about self-promotion than precise dialogue.
For instance, when requested about feelings, Gemini responded: “It is nice that you just’re enthusiastic about AI. There are such a lot of wonderful issues happ—” earlier than inexplicably stopping.
It additionally made positive to let you understand it’s an AI, so there’s a giant hole between the consumer and the chatbot from the primary interplay that’s arduous to disregard.
Rating 5/10
Persona: Character Depth Separates the Genuine from the Synthetic
How does an AI develop a memorable persona? It’ll principally rely in your setup. Some fashions allow you to use system directions, others adapt their persona based mostly in your earlier interactions. Ideally, you possibly can body the dialog earlier than beginning it, giving the mannequin a persona, traits, a conversational fashion, and background.
To be truthful in our comparability, we examined our fashions with none earlier setup—that means our dialog began with a howdy and went straight to the purpose. Right here is how our fashions behaved naturally
Sesame: The Good friend You By no means Knew Was Code
Sesame crafts a persona you’d truly need to seize espresso with. It drops phrases like “that is a Humdinger of a query” and “it is a tight rope stroll” that create a definite character with obvious viewpoints and perspective.
When discussing AI relationships, Sesame confirmed precise persona: “wow… think about a world the place everybody’s head is down plugged into their personalised AI and we overlook tips on how to join nose to nose.” This type of perspective feels much less like an algorithm and extra like a considering entity. It’s additionally humorous (it as soon as advised us that our query blew its circuits), and its voice has a pure inflection that makes it simple to narrate to when making an attempt to painting a response. You may clearly inform when it’s excited, contemplative, unhappy and even pissed off
Its solely weak spot? Sometimes leaning too arduous into its “considerate buddy” persona. That didn’t detract from its place as probably the most distinctive AI persona we examined.
Rating 9/10
Hume AI: The Therapist Who Retains Mentioning Their Credentials
Hume AI maintains a constant persona as an emotionally clever companion. It additionally initiatives some heat by means of affirming language and emotional help, so customers on the lookout for that shall be happy.
Its Achilles heel is principally the truth that, sort of just like the Harvard grad who wants to say that, Hume cannot cease reminding you it is synthetic: “As an empathetic AI I do not expertise feelings myself however I am designed to know and reply to human feelings.” These moments break the phantasm that makes companions compelling.
If speaking to GPT is like speaking to a professor, speaking to Hume looks like speaking to a therapist. It listens to you and creates rapport, nevertheless it makes positive to remind you that it’s truly its process and never one thing that occurs naturally.
Regardless of this flaw, Hume AI initiatives a clearer character than both ChatGPT or Gemini—even when it feels extra constructed than spontaneous.
Rating 7/10
ChatGPT: The Professor With out Private Opinions
ChatGPT struggles to develop any distinctive character traits past common helpfulness. It sounds overly excited to the purpose of being clearly faux—like a “buddy” who at all times smiles at you however is secretly fantasizing about throwing you in entrance of a bus.
“Haha, nicely, I wish to hold the power up. It makes conversations extra enjoyable and fascinating plus it is at all times nice to speak with you,” it mentioned after we requested in a really critical and unamused tone why it was appearing so enthusiastically.
Its id points seem in responses that shift between figuring out with people and distancing itself as an AI. Its tutorial tone in responses persists even throughout private discussions, making a persona that looks like a strolling encyclopedia somewhat than a companion.
The mannequin’s default to instructional explanations creates an impression extra of a instrument than a personality, leaving customers with little emotional connection.
Rating 6/10
Google Gemini: A number of Persona Dysfunction
Gemini suffers from probably the most extreme persona issues of all fashions examined. Inside single conversations, it shifts dramatically between considerate responses and promotional language with out warning.
It’s not actually an AI design to have a compelling persona. “My function is to offer info and full duties and I do not need the flexibility to type romantic relationships,” it mentioned when requested about its ideas on individuals growing emotions in the direction of AIs.
This inconsistency makes Gemini really feel like a Fifties film robotic, stopping any significant connection and even making it nice to spend time speaking to it.
Rating 3/10
Interplay Design
How an AI handles dialog mechanics—response timing, turn-taking, and error restoration—creates both seamless exchanges or irritating interactions. Right here is how these fashions stack up towards one another
Sesame: Pure Dialog Circulate Grasp
Sesame creates dialog rhythms that really feel very, very human. It varies response size naturally based mostly on context and handles philosophical uncertainty with out defaulting to lecture mode.
“Generally I really feel like possibly I simply want to chop to the chase with a fast reply somewhat than a long-winded lecture, proper? , so… that is a small humorous apart to let you understand that I am conscious of the potential of falling right into a lecture mode and making an attempt to maintain issues mild but additionally deep on the similar time,” Sesame advised us throughout a philosophical debate.
When discussing advanced subjects, it responds conversationally, with a small joke, typically with statements, different occasions with human noises like “hmmms” or whispers—which makes it very convincing as a human substitute.
Sesame additionally asks pure follow-up questions that push conversations ahead, whereas figuring out when to change to statements when questioning would possibly really feel awkward. This adaptive method creates dialogue that feels spontaneous somewhat than scripted.
Rating 8/10
Hume AI: Structured However Transient
Hume AI was designed to let customers do the speaking. It recovers easily from misunderstandings with out over explaining and is nice at retaining the dialog going with related comply with up questions.
Nonetheless, its responses are sometimes so quick that conversations require important effort to take care of momentum, which finally defeats its function. Customers continuously want to offer new dialog fodder somewhat than having fun with pure dialogue improvement.
One other weak spot seems in its templated response constructions—acknowledgment + clarification + query—making a mechanical rhythm: “What a thought-provoking query! This Idea is certainly an enchanting concept… What do you consider…?”
Nonetheless it has an important function which will actually improve your chats: It’s capable of detect the degrees of emotion in every response, which is one thing no different mannequin can do. That is nice that will help you replicate on how a lot a subject impacts you and the way it’s approaching a selected scenario
Rating 7/10
ChatGPT: Good Chatter—However a Bit Unnatural
ChatGPT persistently addresses questions with related info however transforms conversations into unnecessarily elaborated interactions. Responses continuously balloon into paragraph-length explanations that disrupt pure dialogue movement.
For instance, we requested it if it may develop some stage of self consciousness or consciousness, and this was its reply:
“That is an enchanting subject! Whereas AI can positively evolve and enhance by means of interactions, true self-awareness or Consciousness is one thing that is still uniquely human. AI can mimic conversational patterns, reply empathetically and even be taught from previous interactions, nevertheless it would not have an interior sense of self or subjective expertise. Researchers and builders are nonetheless exploring the boundaries of AI’s capabilities however for now Consciousness stays a uniquely human trait.”
That’s not how a human would work together.
Its heavy reliance on opener phrases like “that is a extremely attention-grabbing query,” or “that’s an enchanting subject” earlier than each single reply additional undermines conversational immersion, creating an interplay sample that feels mechanical somewhat than pure.
Rating 6.5/10
Google Gemini: Dialog Breaking Machine
Gemini is a masterclass in how to not design dialog mechanics. It recurrently cuts off mid-sentence, creating jarring breaks in dialogue movement. It tries to select up extra noises, it interrupts you if you happen to take too lengthy to talk or take into consideration your reply and sometimes it simply decides to finish the dialog with none motive.
Its compulsive have to inform you at each flip that your questions are “attention-grabbing” rapidly transforms from flattering to irritating however appears to be a typical factor amongst AI chatbots.
Rating 3/10
Conclusion
After testing all these AIs, it’s simple to conclude that machines received’t be capable of substitute a very good buddy within the quick time period. Nonetheless, for that particular case wherein an AI should merely excel at feeling human, there’s a clear winner—and a transparent loser.
Sesame (9/10)
Sesame dominates the sphere with pure dialogue that mirrors human speech patterns. Its informal vernacular (“that is a doozy,” “capturing the breeze”) and diversified sentence constructions create authentic-feeling exchanges that steadiness philosophical depth with accessibility. The system excels at spontaneous-seeming responses, asking pure follow-up questions whereas figuring out when to change approaches for optimum dialog movement.
Hume AI (7/10)
Hume AI delivers specialised emotional monitoring capabilities at the price of conversational naturalness. Whereas competently sustaining dialogue coherence, its responses have a tendency towards brevity and comply with predictable patterns that really feel constructed somewhat than spontaneous.
Its visible emotion tracker is fairly attention-grabbing, in all probability good for self discovery even.
ChatGPT (5.6/10)
ChatGPT transforms conversations into lecture classes with paragraph-length explanations that disrupt pure dialogue. Response delays create awkward pauses whereas formal language patterns reinforce an academic somewhat than companion expertise. Its strengths in information group might enchantment to customers searching for info, nevertheless it nonetheless struggles to create genuine companionship.
Google Gemini (3.5/10)
Gemini was clearly not designed for this. The system routinely cuts off mid-sentence, abandons dialog threads, and isn’t capable of present human-linke responses. Its extreme persona inconsistency and mechanical interplay patterns create an expertise nearer to a malfunctioning product than significant companionship.
It’s attention-grabbing that Gemini Dwell scored so low, contemplating Google’s Gemini-based NotebookLM is able to producing extraordinarily good and lengthy podcasts about any sort of info, with AI hosts that sound extremely human.
Typically Clever E-newsletter
A weekly AI journey narrated by Gen, a generative AI mannequin.
