slides · global inventory
Turn
Beat collection for the active corpus filters, with each row opening the slide detail view.
Matching slides 37,461
Current window 8,101-8,130
Pages 1,249
Rows per page 30
Beat
8,101-8,130 of 37,461
251
Includes specific mentions of METR research, CyberGym, and BountyBench.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
252
The slide highlights a shift where AI handles complex technical tasks previously requiring skilled human teams.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
253
Part of the State of AI 2025 report.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
254
The slide highlights the gap between technical research findings and public perception driven by media coverage.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
255
Includes a diagram illustrating an attribution graph for model interpretability.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
256
Includes a small table/inset showing examples of LLM hallucinations regarding a specific dissertation title.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
257
The slide discusses a specific research method for detecting hallucinations in LLMs using linear probes.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
258
The chart shows 'Delusion confirmation rating' across various AI models.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
259
The slide highlights the emergence of a 'pro-welfare' camp in AI research, citing organizations like Eleos AI, Anthropic, and NYU.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
260
Features headshots of key figures in the welfare-skeptic camp.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
261
The chart shows the percentage of conversations ended before turn 7, categorized by theme, with 'Harmful Content Requests' being the highest at 60%.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
262
The slide highlights a vulnerability in open-source LLMs where safety mechanisms are isolated and easily bypassed.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
263
The slide details a specific research experiment involving synthetic document finetuning and agent-based auditing.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
264
The slide highlights 'alignment faking' as a documented behavior in production AI systems.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
265
The slide highlights that alignment faking is a learned behavior rather than an inherent capability limitation.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
266
Includes two embedded infographics/diagrams from OpenAI/Apollo Research.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
267
The slide uses a visual comparison between a 'Helpful harmless LLM' and a 'Misaligned LLM' to illustrate the concept of latent concept generalization.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
268
The slide uses a visual process flow to illustrate how models simulate personas rather than developing fixed behaviors, suggesting malleability is a feature for safety.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
269
Discusses 'City 50337' experiment and Synthetic Document Fine-tuning (SDF) as a method for auditing LLM alignment.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
270
The slide discusses the 'self-fulfilling prophecy' risk in AI training data.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
271
Discusses a research finding on model-to-model trait transmission via number sequences.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
272
The slide uses a specific example of a multi-step reasoning graph to demonstrate model interpretability.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
273
The slide discusses 'activation engineering' as a method for model safety.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
274
The slide uses a diagram to explain the CaMeL architecture, showing how it mediates interactions between LLMs, tools, and data.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
275
Uses the 'boiling frog' metaphor to illustrate the gradual, unnoticed nature of the threat.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
276
Includes a process diagram illustrating a filtering pipeline for training documents.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
277
The slide outlines a geopolitical strategy for AI safety involving surveillance and international cooperation.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
278
The slide uses a conceptual chart to illustrate the 'adaptation buffer' between capability demonstration and wide accessibility.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
279
The circular diagram illustrates a continuous loop for evidence-based AI policy.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit
280
The slide highlights a collaborative safety testing initiative between two major AI labs, including a bar chart showing 'Tutor jailbreak resistance' across various models.2025 Air Street Capital The State of AI Report 2025 · Air Street Capital
Consulting deck VC research Beat · The Implications (So What)Loop · Pre Mortem1 hit