Alex Prompter

This paper from Harvard and MIT quietly answers the most important AI question nobody benchmarks properly:

Can LLMs actually discover science, or are they just good at talking about it?

The paper is called “Evaluating Large Language Models in Scientific Discovery”, and instead

Most AI benchmarks test answers.

This paper tests the process of discovery.

Models must:

• Form hypotheses
• Design experiments
• Observe outcomes
• Update beliefs
• Repeat under uncertainty

That’s real science, not Q&amp;A.

LLMs are surprisingly good at the first step.

They generate plausible, well-worded hypotheses that look exactly like something a researcher would write.

But that’s where the illusion starts.

Once experiments produce results, models struggle.

They:

• Ignore contradictory evidence
• Stick to wrong theories
• Rationalize failures instead of correcting them

They don’t update they defend.

A major failure mode:

Models latch onto correlations and invent causal stories.

Even when experiments explicitly break the link, the explanation doesn’t change.

Fluency replaces falsification.

Here’s the uncomfortable result:

High scores on standard reasoning benchmarks
≠
High scores on scientific discovery tasks

The correlation is weak or nonexistent.

LLMs can talk science.

They can’t yet:

• Track hypotheses over time
• Revise beliefs reliably
• Learn from failed experiments

Scientific intelligence is iterative, not eloquent.

Until models learn that, “AI scientist” is still a metaphor.

Read full paper here:

AI prompts your competitors don't want you to find

→ Prompts for marketing, sales, operations
→ Unlimited custom prompts
→ Pay once, own forever

Grab it today 👇

the day they find out how to make them think and discover new science; we will get AGI.

yess, so many teams from around the globe collaborated

10x your prompting skills with my prompt engineering guide

→ Mini-course
→ Free resources
→ Tips &amp; tricks

Grab it while it's free ↓

OpenAI, Anthropic, and Google AI engineers use 10 internal prompting techniques that guarantee near-perfect accuracy…and nobody outside the labs is supposed to know them.

Here are 10 of them (Save this for later):

Technique 1: Role-Based Constraint Prompting

The expert don't just ask AI to "write code." They assign expert roles with specific constraints.

Template:

You are a [specific role] with [X years] experience in [domain].
Your task: [specific task]
Constraints: [list 3-5 specific

Technique 2: Chain-of-Verification (CoVe)

Google's research team uses this to eliminate hallucinations.

The model generates an answer, then generates verification questions, answers them, and refines the original response.

Template:

Task: [your question]

Step 1: Provide your

Technique 3: Few-Shot with Negative Examples

Anthropic discovered that showing the model what NOT to do is as powerful as showing what TO do.

Template:

I need you to [task]. Here are examples:

✅ GOOD Example 1: [example]
✅ GOOD Example 2: [example]

❌ BAD Example 1:

Technique 4: Structured Thinking Protocol

OpenAI's GPT-5 team uses this for complex reasoning tasks.

Force the model to think in layers before responding.

Template:

Before answering, complete these steps:

[UNDERSTAND]
- Restate the problem in your own words
- Identify what's

Technique 5: Confidence-Weighted Prompting

Google DeepMind uses this for high-stakes decisions.

Ask the model to rate its confidence and provide alternative answers.

Template:

Answer this question: [question]

For your answer, provide:
1. Your primary answer
2. Confidence

Technique 6: Context Injection with Boundaries

Anthropic engineers inject massive context but set clear boundaries on what matters.

Template:

[CONTEXT]
[paste your documentation, code, research paper]

[FOCUS]
Only use information from CONTEXT to answer. If the answer isn't in

Technique 7: Iterative Refinement Loop

OpenAI's research team chains prompts to refine outputs through multiple passes.

Template:

[ITERATION 1]
Create a [draft/outline/initial version] of [task]

[ITERATION 2]
Review the above output. Identify 3 weaknesses or gaps.

[ITERATION

Technique 8: Constraint-First Prompting

Google Brain researchers start with constraints before the actual task.

Template:

HARD CONSTRAINTS (cannot be violated):
- [constraint 1]
- [constraint 2]
- [constraint 3]

SOFT PREFERENCES (optimize for these):
- [preference 1]
-

Technique 9: Multi-Perspective Prompting

Anthropic's Constitutional AI uses multiple viewpoints to reduce bias and improve reasoning.

Template:

Analyze [topic/problem] from these perspectives:

[PERSPECTIVE 1: Technical Feasibility]
[specific lens]

[PERSPECTIVE 2: Business

Technique 10: Meta-Prompting (The Nuclear Option)

This is what OpenAI's red team uses to break their own models and find edge cases.

You ask the AI to generate the perfect prompt for itself.

Template:

I need to accomplish: [high-level goal]

Your task:
1. Analyze what would

You can use these techniques RIGHT NOW

I've been using these with Claude Sonnet 4.5, GPT-4, and Gemini 2.0 Flash for 6 months.

Results:

- 100% reduction in hallucinations on technical docs
- 3x faster iteration on code generation
- 90%+ accuracy on complex analysis tasks

The

I hope you've found this thread helpful.

Follow me @alex_prompter for more.

Like/Repost the quote below if you can: