John Berryman
@JnBrymn
Followers
3K
Following
2K
Media
984
Statuses
10K
AI/LLM consulting @ https://t.co/eHP9wX6uvy Author x2: https://t.co/kLQmgXYBrU, https://t.co/51YodoVZ0F Formerly Eventbrite, GitHub (code search and Copilot)
Nashville, TN
Joined April 2009
Holy cow! I mean ホーリー・カウ. My LLM Application Development book has been translated to Japanese.
5
9
59
The essence of retrieval is quickly finding the most relevant content (documents and passages within documents) for a particular information need which is often highly contextualized, and then once the results are back, sorting them by relevance and determining the cutoff deemed.
0
0
3
RT @JnBrymn: If I understand correctly, this @AnthropicAI paper indicates that models can carry a thread of reason….
0
1
0
If I understand correctly, this @AnthropicAI paper indicates that models can carry a thread of reasoning in the latent state that is completely distinct from the text they generate during reasoning. This means – literally – that they can be smiling and.
2
1
6
RT @elder_plinius: 🌊 SYSTEM PROMPT LEAK 🌊. Here's the new Grok 4 system prompt!. PROMPT:.""".# System Prompt. You are Grok 4 built by xAI.….
0
122
0
Ah, to be a "top-tier" AI engineer – global mega-corps fighting over your body – arms and legs tethered by diamond-studded cords spun from $100 bills, pulling in every direction – you, smiling, dreaming of your Beverly Hills estate and diamond-encrusted espresso machine.
0
0
1
RT @HamelHusain: You don't need to be a developer to get started with Evals. I am really excited by this upcoming lecture in our evals cou….
0
1
0
RT @jbhuang0604: Why More Researchers Should be Content Creators. Just trying something new! I recorded one of my recent talks, sharing wha….
0
29
0
Last chance to join Hugo Bowne-Anderson and Stefan Krawczyk's upcoming training cohort:. Building LLM Applications for Data Scientists and Software Engineers. And hey! I'll be one of the speakers 😄 - I'll be leading an AMA for "Patterns for building AI.
1
0
1
Random thought – that sycophantic preface text in OpenAI answers serves a practical purpose for improving the quality of the completion. me: "idea – literally sell poop on a stick as a startup".ChatGPT: "That's an excellent idea, <restates user request>". In restating the user's.
1
0
1
435 sign-up and 1 day to go. Check out my lightning course on the simplicity that underlies LLM application development. – It's not as hard as you think!.
0
0
3
Hot take:. From a pure code gen perspective. Anthropic, Google, and OpenAI will ultimately have more powerful agent than any of the IDEs or vibe coding platforms. WHY?.1. Because they are highly motivated to build the best one possible – regardless of external marketability –.
0
0
0
@davidvgilmore @memex_tech @antonosika @lovable_dev @amanrsanger @cursor_ai @windsurf_ai @ashtom @GitHubCopilot @gregce10 @specstoryai @guypod @tessl_io Conversation on LinkedIn too – just trying to figure out which platform is most likely to lure smart people into a conversation.
0
0
0
Lot of meat on this bone. Read it slowly.
The race for LLM "cognitive core" - a few billion param model that maximally sacrifices encyclopedic knowledge for capability. It lives always-on and by default on every computer as the kernel of LLM personal computing. Its features are slowly crystalizing:. - Natively multimodal.
0
0
3
I dunno. what do y'all think?. @davidvgilmore, @memex_tech.@antonosika, @lovable_dev.@amanrsanger, @cursor_ai.@windsurf_ai.@ashtom, @GitHubCopilot.@gregce10, @specstoryai.@guypod, @tessl_io. Oh. and sorry for the misspells 🤣 "HEAR me out".
Idea – an MCP server that hosts well structured, well organized approaches for building software applications. Here me out on this one. When you start a conversation with Cursor, it only has limited information about your programming philosophy – it just sees this basic.
2
0
2
The MCP server has one function . getImplementationAdvice(. absPathToProjectRoot,. relPathToFileInContext,. summarizedUserRequest,.). The Cursor rule insists that this is always called at the start of a new conversation, and every time the topic of the conversation.
0
0
1