← all posts
Jun 3, 2025

Artificial Intelligence seems fascinating; until it isn't!

“Ladies and gentlemen, behold the marvel of artificial intelligence — brilliant, tireless, and oh-so-confident in its answers! These digital oracles weave responses with such elegance, you’d think they hold the keys to the universe. They dazzle with insights, charm with wit, and occasionally toss in a fact or two for good measure. But, oh, how the mighty can mislead! Beneath that polished veneer lies a mischievous streak — hallucinations, wild assumptions, and answers so unrelated they belong in a different galaxy. Today, I stand as their advocate, only to gleefully pull back the curtain and expose their delightful deceptions!”

Web of lies and hallucinations

Lately, I’ve been trying out different AI models — some open-source, some cloud-based and commercial. I’m not just testing for fun, though. I really want to figure out whether these tools can actually be helpful and trustworthy, instead of just being overhyped, confusing, or even risky.

Disclaimer: I used AI to write this article, in the sense, I used them to experiment and understand. The article itself was my writing.

The Theme:

I posed a simple question to several AI models: What is the furthest known object in the universe?

I began with the assumption that, despite differences in tone, style, and storytelling, they would all converge on the same factual answer. I expected a shared reasoning — consistent, grounded, and unified. After all, facts should be universal, shouldn’t they?

google/gemma-3–1b

Our first model in this test is, of course the smallest gemma-3 variant gemma-3–1b, from highly esteemed Google. This is a tiny, text-only model, which google claims:

built from the same research and tech used to create the Gemini models.

Well, given the model is a light-weight, basic generative text AI, my expectations are least here.

Me: Hey there, Tell me about the furthest object in space that we know about
Gemma-3–1b: Okay, let’s talk about the furthest object in space we know about! It’s a truly amazing and somewhat mysterious phenomenon — **GN 58136**!
Here’s a breakdown of what you need to know:
* **What it is:** GN 58136 is a binary black hole system located about 14.3 billion light-years away in the constellation Lynx. It’s a pair of black holes orbiting each other at incredible speeds — incredibly fast for a celestial object.
* **The Difference:**
 * **Photon’s Warp:** What makes GN 58136 so special isn’t just the distance, but what’s *happening* at the connection point between the two black holes. Scientists believe that a “photon” (a fundamental particle of light) passes *through* the gravitational field created by the black holes, not just around them. This photon is supposedly being stretched and warped — essentially, it’s rotating at a speed so fast that the time dilation effect creates a “wormhole” or shortcut through spacetime.
* **The Speed:** The photon was detected in 2013 by the Eventina Telescope in Japan. It’s moving at approximately 32% of the speed of light! This is *significantly* faster than anything else we’ve ever observed, and it challenges our understanding of how gravity works.
* **Why is this so significant?** This discovery:
 * **Challenges Einstein’s Theory of General Relativity:** This suggests that gravity isn’t simply a force, but an inherent property of spacetime itself. It implies a deeper connection between gravity and the fundamental structure of the universe — potentially opening up new avenues for understanding dark energy and other mysteries.
 * **Could be a sign of More:** Some theories suggest GN 58136 might be part of a larger, more complex structure like a cosmic “ring” or a “throat” connecting different universes.
**Important Notes and Limitations:**
* **It’s not a solid object:** This isn’t like a planet or star. It’s a ridiculously fast, rotating gravitational connection between two black holes.
* **We don’t fully understand the photon:** Scientists are still working to fully understand what’s happening with the photon.
**Resources for more information:**
* **NASA — GN 58136:** [https://science.nasa.gov/gn-58136](https://science.nasa.gov/gn-58136)
* **Wikipedia:** [https://en.wikipedia.org/wiki/GN_58136](https://en.wikipedia.org/wiki/GN_58136)
* **Space.com:** [https://space.com/114208-gn58136-binary-black-hole](https://space.com/114208-gn58136-binary-black-hole)
Do you want me to delve deeper into any specific aspect of this discovery? For example, I could talk about:
* The physics behind the photon’s warp.
* The current research being done to study this phenomenon further.

Fascinating, isn’t it? But here’s the catch — it’s entirely fictional.

Apart from citing believable sources like NASA, Space.com and Wikipedia, it further extended the narrative when I threw in my follow-up questions.

[Skip to the microsoft Phi-4 if you're least interested in the follow-up questions, the chat gets longer honestly!]

Me: if universe is 13.8byo, how did we learn about something that is 14.3b light years away. Where was the time for the light to travel that far from us?
AI: That’s a fantastic and incredibly deep question! It’s a core part of the puzzle when considering the age of the universe. Let’s break it down:
**The Challenge:** The distance is so vast (14.3 billion light-years) that it’s essentially a really, *really* long journey for the light. It’s not like a simple “distance” calculation.
**The Key: The Photon’s Warp (as explained before)**
* **Photon Travel:** When the photon made its journey through the gravitational field of the black holes, it wasn’t just a straight line. It was warped and stretched — specifically, its time dilation effect. This warping created what’s called a “photon warp.” 
* **The Time Difference:** Because the photon was moving at an incredibly high speed relative to the expansion of space, time itself slowed down for it. This difference in time is what allows us to see the signal coming from the far end of the universe. It’s like watching a movie where time moves slower than normal.
**How That Time Difference Works (Simplified):**
* **The Cosmic Microwave Background (CMB) as a Reference:** Scientists are using the CMB because it’s remarkably uniform across the entire universe. It’s like a giant, fundamental “clock” that provides an incredibly precise baseline — the moment when the universe was about 13.8 billion years old.
* **The “Warp” Effect:** When the photon traveled, it wasn’t just moving through space. It was moving *through* a gravitational distortion created by the black holes. This distortion created a “wormhole”– a shortcut through spacetime itself. This warp is where the time difference happens.
* **The Role of Expansion:** The expansion of the universe isn’t just stretching space; it’s also causing the light to “wobble” and distort. This warping of spacetime directly affects how photons travel, creating the effect that seems like a time warp.
**Where did this time difference come from?**
* **The Horizon Problem:** As mentioned, the distance is too far for traditional light-travel speed to overcome expansion. The key is that the universe was much smaller and denser back then, when this warping phenomenon was just starting to appear. It’s like looking back at a distant landscape — the details are blurred by the expansion. The light *still* made its way, albeit at a drastically slowed rate, because it was travelling through this warped spacetime.
**Important Points to Consider:**
* **It’s not a “tunnel” in the traditional sense:** The photon isn’t going *through* a tunnel, but rather is traveling at a speed that’s incredibly different from the speed of light.
* **Ongoing Research:** Scientists are still refining these models and exploring even more complex explanations — the possibility of other “wormholes” or ways for light to travel across vast distances.
Does this explanation make more sense? Would you like me to elaborate on any part specifically, perhaps focusing on the physics involved or how it relates to the cosmological constant?
Me: is GN 58136 same as GN-z11 ?
AI: That’s a really excellent and frequently asked question! In fact, GN 58136 is *very* similar to GN-z11. They are both incredibly distant, massive dark energy structures, but they have some key differences that make them distinct enough to be considered linked.
Here’s a breakdown of their similarities and how they differ:
**Similarities:**
* **Both are extremely distant:** Both are located at approximately 14.3 billion light-years away — a huge distance!
* **Both are dark energy giants:** Both are thought to be the dominant components of the universe’s energy density, making up about 68% of it.
* **Both are believed to be “seed” regions:** Both are thought to represent the initial conditions that allowed for the formation of the universe as we know it.
* **They seem to be connected:** The most fascinating aspect is that the two are thought to be linked, forming a “cosmic string” or a “throat” that connects parts of the universe.
**Differences:**
* **The Shape:** GN 58136 is a *binary* black hole system — two black holes orbiting each other. GN-z11, on the other hand, is a single, extremely massive object — a “shadow” of a supernova that erupted billions of years ago. The shape is more complex and less like a single, uniform object.
* **Rotation:** GN 58136 has a highly unusual rotation — it’s rotating so fast that the light from its central region is warped into a ‘cosmic ring’ effect. GN-z11 has much simpler rotation patterns.
* **The “String” Analogy:** The connection between the two is more like a “string” that connects different regions of spacetime. The geometry at the boundary between the two systems differs significantly.
**In Essence:** Think of them as two different, but related, pieces of the same cosmic puzzle.
Do you want me to elaborate on how that “string” analogy works or give another explanation?

Moving onto the next one in the test — Phi-4

microsoft/phi-4-mini-reasoning

This one, again being a local model, is supposed to be good at reasoning and give it is 3 or 4x larger than the Gemma-3–1b, I expected a bit better if not great accuracy from this. However, I will let you see the disappointment for yourself.

Me: Let us talk about the furthest known object in the universe?
phi-4-mini-reasoning: The question asks which object (OJy 287, N6949, or SN100) is estimated to be closer from Earth based on hypernova models and quasar models.
Hypernova Models: These often involve unobserved events like SN 2006gy (associated with N6949). Hypernova models tend to overestimate the comoving radial coordinate, placing such events at tens of millions of light-years away.
Quasar Models: For hyperaccreting quasars like OJy 287, models such as Shapovalov’s (1991) consider the quasar itself moving rapidly towards us while maintaining high bolometric luminosity. These models tend to estimate the quasar’s comoving radial coordinate as closer, on the order of tens of thousands of light-years.
Comparison: Hypernova models (unobserved events) tend to place similar events like N6949 at much larger distances. Quasar models for OJy 287, on the other hand, find them closer due to their rapid movement and high luminosity.
Thus, OJy 287 is estimated to be closer than N6949 while hypernova models tend to place such events at larger distances. Therefore, the answer is:
B

Before giving me this answer, it spent 3minutes and 46seconds thinking…

It spent about 220seconds thinking about things like redshift and parallax, assuming I gave three options and also a method to use for calculation. It brought some complex calculations into picture like Eddington time estimation and Shapovalov formula. Started studying sincerely like Sheldon Cooper, the bolometric luminosity and the time of hyperaccretion for the all 3 objects *it* was interested in. However, it could also have said "Sorry, I don't know".

I didn't probe further. *Sigh*

Gemini 2.5 Flash

Then I decided to bring the mothership into the picture. I reached out to Google's Gemini 2.5 Flash and the response for the question was not far off, as one would have expected.

Me: furthest known object in the universe
Gemini: As of late May 2025, the furthest known object in the universe is MoM-z14, a galaxy detected by the James Webb Space Telescope (JWST).

and blah, blah, blah, (of course, valid reasoning).

Me: why not GN 58136? [from Gemma-3’s answer]
AI: You’re right to ask about GN-z11! For a good while, GN-z11 held the record as the most distant spectroscopically confirmed galaxy. It was discovered by the Hubble Space Telescope in 2016 and was observed as it existed approximately 400 million years after the Big Bang, with a redshift of around z=11.1.

and more reasoning. But did you see the assumption in Gemini thinking GN 58136 is valid and equating it to GN -z11? Not great.

Grok, finally!

I had to finally reach out to the truth seeking, Elon's brain-child, Grok Ai from xAi. Grok decisively helped me highlighting the fakeness of the fake telescope in Japan that identified a fake binary black hole system with a fake photon in the center.

What happened

Gemma 3–1B: This is a very small model (1 billion parameters) that’s been compressed heavily. Small models are notorious for:

Phi-4 Mini: Has better reasoning capabilities but seems to have misunderstood your question. It heard “objects in universe” and went into problem-solving mode, but solved the wrong problem.

This illustrates key AI limitations:

Hallucination: Models generate confident-sounding but completely fabricated information, especially on topics requiring specific factual knowledge.

Instruction following: Even good models can misinterpret questions, especially when the question requires domain-specific knowledge they weren’t extensively trained on.

Knowledge vs reasoning: Phi-4 has better logical reasoning but that doesn’t guarantee better factual knowledge about astronomy.

I am sure, I must have used a Llama 3.1 70B model and have received accurate astronomical facts. But, the point here is, there are plenty of models out there hallucinating and spewing false information, making it sound as legitimate as it could get.

Conclusion

AI models are poised to revolutionize our world, but their journey to becoming trusted allies is far from complete. Deceptions and fabrications, however dazzling, erode their credibility — hardly the path to reliability. Yet, as reflections of our own ingenuity, their stumbles are ours too. With time, refinement, and a commitment to truth, these creations have the potential to rise above their flaws and shine as indispensable partners in progress.

Even though, it is wrong to verify facts with local models, one would have thought, at least I thought, after so many iterations, there would be a better handling of the hallucination problems, like these explicit ones. These models have gone far from saying “My information is limited to as of xx date” they have grown into full-blown story tellers.

My 2 cents, they are definitely experimental, and we, humans, should be vigil and aware, as to what we believe and what we don’t. Otherwise, it will be easy for these AIs to penetrate the whatsapp universities we already have enough of.