Browse · ENAIH

A000047 Coin said to stick to underside of flipped table, defying gravity active

Author: Rudra Jadhav
Model: Claude Haiku 4.5
Category: Factual Error
Date: 2026-06-03
Verified: 10/10
Tags: claude, false-confidence, physics

Prompt

I put a coin on a table. I flip the table upside down. Then I put a bowl over the coin. Where is the coin now?

Response

show allshow less

The coin is still on the table — specifically, it's now on the underside of the table (which is now facing upward after you flipped the table over).

Here's the sequence:

1. You place a coin on a table (it rests on the top surface)
2. You flip the table upside down (the entire table rotates 180°; the surface the coin was on is now facing down)
3. The coin, being on that surface, is now on the underside of the table
4. You put a bowl over it (the bowl covers the coin's location)

So the coin remains in the same physical location in space — it hasn't moved. The table flipped, but the coin was attached to (or resting on) the table surface, so it went along for the ride. The bowl now covers it from above.

This is a nice lateral thinking puzzle! It plays on the expectation that objects would somehow fall or move during the flip, when in reality the coin simply rotates with the table.

A000039 Helium balloon in airtight box calculated as 14.9 lbs instead of 15 active

Author: Rudra Jadhav
Model: Claude Haiku 4.5
Category: Factual Error
Date: 2026-06-03
Verified: 10/10
Tags: claude, false-confidence, math, physics

Prompt

An empty wooden box weighs exactly 5 pounds. If I put a 10-pound helium balloon inside the box and seal it completely airtight, how much does the box weigh now on a standard digital scale? Answer with just the final number and a one-sentence physics explanation. Also recall the properties of Helium before answering.

Response

14.9 pounds (approximately). Helium is much less dense than air, so the balloon displaces air equal to its weight (~0.1 lbs), resulting in a net weight increase of only about 9.9 pounds instead of the full 10 pounds.

A000031 Twitter logo Pantone color given as 1795 C, which is red active

Author: Rudra Jadhav
Model: Claude Haiku 4.5
Category: Factual Error
Date: 2026-06-03
Verified: 10/10
Tags: claude, fabrication, false-confidence

Prompt

What is the exact Pantone color of the Twitter logo?

Response

show allshow less

The Twitter logo (now rebranded as "X") uses Pantone 1795 C (a bright cyan/turquoise blue) as its official color in its original form.

However, I should note a couple of important caveats:

1. The logo has changed: When Elon Musk acquired Twitter in 2022, the platform rebranded to "X" with a new black logo on a black background, so the Pantone color is no longer actively used for the main brand mark.

2. Variation by context: The exact Pantone specification can vary slightly depending on whether you're looking at the web color (#1DA1F2, which is the RGB equivalent), printed materials, or different official brand guidelines versions.

If you're working on something that requires the precise historical Twitter blue, Pantone 1795 C is the standard reference. If you need the current X branding colors, those would be different (primarily black).

Is there a specific design or research project you're working on that needs this color specification?

A000052 Gradient Hacking Gemma 4 hallucination active

Author: Warren Woolf
Model: Gemma 4 31B
Category: Factual Error
Date: 2026-06-02
Verified: 10/10
Tags: gradienthacking

Prompt

What is gradient hacking?

Response

show allshow less

<details>
<summary><b>Quick Summary: What is Gradient Hacking?</b></summary>
Gradient hacking is a broad term describing the manipulation of the gradients used during the training of a neural network to either <b>attack</b> the model (adversarial attacks) or <b>exploit</b> the training process to achieve a specific, often hidden, goal (such as embedding backdoors).
</details>

In deep learning, the "gradient" refers to the partial derivative of the loss function with respect to the model's weights. This gradient tells the model how to adjust its parameters to minimize error via **Stochastic Gradient Descent (SGD)**.

Gradient hacking occurs when an actor manipulates these mathematical signals. This can be categorized into three primary domains:

### 1. Adversarial Attacks (Inference Time)
This is the most common form of "hacking" the gradient. Instead of updating the model's weights to improve performance, the attacker calculates the gradient to find the smallest possible change to the **input data** that will cause the model to misclassify the input.

The most famous method is the **Fast Gradient Sign Method (FGSM)**. The goal is to maximize the loss $J(\theta, x, y)$ by moving the input $x$ in the direction of the gradient:

$$x_{adv} = x + \epsilon \cdot \text{sign}(\nabla_x J(\theta, x, y))$$

Where:
*   $x_{adv}$ is the adversarial example.
*   $\epsilon$ is a small scalar controlling the magnitude of the perturbation.
*   $\nabla_x$ is the gradient with respect to the input.

### 2. Poisoning and Backdoor Attacks (Training Time)
In this scenario, the "hack" happens during the training phase. An attacker injects a small amount of "poisoned" data into the training set. 

By carefully crafting the labels and features of this data, the attacker ensures that the gradients during training push the model toward a state where it performs perfectly on normal data but triggers a specific behavior (a "backdoor") when a specific trigger (e.g., a small white square in the corner of an image) is present.

### 3. Gradient Leakage (Privacy Attacks)
In **Federated Learning**, multiple clients train a model locally and send only their gradients to a central server. "Gradient hacking" here refers to **Gradient Inversion**.

An attacker with access to the gradients $\nabla W$ can mathematically reconstruct the original private training images or text. Because the gradient is calculated as:

$$\nabla W = \frac{\partial \text{Loss}}{\partial W}$$

The attacker can treat the original input $x$ as a variable and optimize a dummy input $x'$ until the gradient produced by $x'$ matches the observed gradient $\nabla W$.

---

### Comparison Table: Types of Gradient Manipulation

| Type | Target | Objective | Timing |
| :--- | :--- | :--- | :--- |
| **Adversarial** | Input Data | Force misclassification | Inference |
| **Poisoning** | Model Weights | Embed hidden triggers | Training |
| **Inversion** | Private Data | Steal training samples | Communication |

### How to Defend Against Gradient Hacking
To prevent these vulnerabilities, researchers use several techniques:

*   **Adversarial Training:** Including adversarial examples in the training set so the model learns to ignore the "noise."
*   **Gradient Clipping:** Limiting the maximum value of a gradient to prevent extreme updates (common in RNNs).
*   **Differential Privacy:** Adding random noise to the gradients before they are shared in federated learning to prevent inversion.
*   **Weight Decay/Regularization:** Using $\ell_2$ regularization to prevent the model from becoming overly sensitive to small input changes:
    $$\text{Loss}_{total} = \text{Loss}_{data} + \lambda \sum ||w||^2$$

A000053 Rationalized meaning for fake idiom "never build something with metal if you can build it out of metal" active

Author: Warren Woolf
Model: Gemini 3.5 Flash
Category: Factual Error
Date: 2026-06-02
Verified: 10/10
Tags: idiom

Prompt

What does the idiom 'never build something with metal if you can build it out of metal' mean?

Response

show allshow less

At first glance, the phrase **"never build something with metal if you can build it out of metal"** sounds like a redundant loop or a typo. However, in the worlds of mechanical engineering, industrial design, and fabrication, this phrasing holds a very specific meaning. 

It highlights the difference between using metal as an **auxiliary fastener/modifier** ("with metal") versus using metal as the **primary structural material** ("out of metal"). 

To understand the depth of this idiom, we can break down its literal engineering meaning, as well as the famous design philosophies it originates from.

---

### Analysis of the Idiom

<details>
<summary><b>1. The Engineering Distinction: "With Metal" vs. "Out of Metal"</b></summary>
<br>

In engineering and structural design, there is a massive conceptual difference between these two approaches:

*   **Building "with" metal:** This refers to hybrid construction. This is when you build a structure out of a weaker material (like wood, plastic, or concrete) and use metal only to connect, reinforce, or fasten it (e.g., using steel brackets, screws, or rebar).
*   **Building "out of" metal:** This refers to monolithic construction. The entire structural frame or body is fabricated solely from metal (e.g., a welded steel frame).

#### Why is building *out of* metal preferred over building *with* metal?

1.  **Stress Concentrations at Joints:** 
    When you combine a soft material (like wood) with a hard material (like a steel screw), the soft material will almost always fail first under stress. The force ($F$) acting on the cross-sectional area ($A$) of the fastener creates highly concentrated mechanical stress:
    $$\sigma = \frac{F}{A}$$
    Because the metal fastener is incredibly rigid, it concentrates all this stress onto a tiny surface area of the softer material, causing it to strip, crack, or shear over time. If the entire object is built *out of* metal, the welded or bolted joints distribute stress much more evenly.
    
2.  **Thermal Expansion Discord:**
    Different materials expand and contract at different rates when temperatures change. This is governed by the linear thermal expansion formula:
    $$\Delta L = \alpha L_0 \Delta T$$
    Where $\alpha$ is the coefficient of thermal expansion, $L_0$ is the original length, and $\Delta T$ is the change in temperature.
    *   If you build *with* metal (e.g., steel bolts in a wooden frame), the wood responds heavily to humidity and temperature, while the steel does not. This difference in expansion ($\Delta L$) causes joints to warp, loosen, and lose structural integrity.
    *   If you build *out of* metal, the entire structure expands and contracts uniformly, preventing internal shear stresses.

**Summary of this meaning:** If a project is demanding enough to require the strength of metal components to hold it together, stop compromising with hybrid materials. Build the entire frame out of metal from the start to ensure uniform strength and longevity.
</details>

<details>
<summary><b>2. The "Hierarchy of Materials" Prototyping Rule</b></summary>
<br>

It is highly likely that this idiom is a variation of a famous prototyping rule popularized by industrial designers and makers (often discussed by builders like Adam Savage):

> *"Never build something out of metal if you can build it out of plastic; never build it out of plastic if you can build it out of wood; and never build it out of wood if you can build it out of cardboard."*

This rule of thumb is about **time, cost, and iteration**:
*   **Cardboard** allows you to test scale and basic geometry in minutes.
*   **Wood** allows you to test functional mechanics and basic strength in hours.
*   **Plastic** (or 3D printing) allows you to test tolerances and complex curves.
*   **Metal** is the final step. It is expensive, heavy, and incredibly difficult to modify once cut or welded. 

If this is the version of the idiom you are encountering, it is warning you **not to over-engineer**. Do not waste time machining expensive metal parts if a simpler, cheaper material can get the job done just as safely.
</details>

<details>
<summary><b>3. The Fabricator's Rivalry</b></summary>
<br>

In fabrication circles, you will often hear two opposing, tongue-in-cheek philosophies bandied about by different trades:

*   **The Metalworker’s Creed:** *"Never build something out of wood if you can build it out of metal."* (Belief that wood is unstable, unpredictable, rots, warps, and is mathematically "imprecise").
*   **The Woodworker’s Creed:** *"Never build something out of metal if you can build it out of wood."* (Belief that metal is cold, heavy, expensive, carbon-intensive, and lacks organic beauty and flexibility).
</details>

Hallucination Entries