Alex Xi's Notes
Explorer
Writings
Debugging Issues from Postgres JSONB
Engineering
AI Team Org Map (April 2026)
AI Team Org Map (April 2026).sync-conflict-20260406-161943-6HXDDG5
AI Team Org Map (April 2026).sync-conflict-20260406-162533-6HXDDG5
Framework for Technical Decisions
ML
GPU Reading Group
Reading Group Plan
Session 1
AI
day of session guide
email templates
exercise code files
gpu study session1 plan
host preparation
quick prep checklist
quick reference card
session backup plans
session materials overview
Agenda
Prep
Prereqs
Session 2
Prep
Notes On Finetuning
My experience with GPU vendors
On scaling an engineering stack (WIP)
Startup Engineering Growth Stages
To the interviewees
Mindset
Active Thinking and Processes (WIP)
Random
Startups
Capturing Value From Technology Diffusion (WIP)
On Serverless Providers (WIP)
On startup velocity (WIP)
Templates
Daily Note Template
Weekly Note Template
Weekend Projects
Bardock - a Monorepo Dockerfile management tool
Building a secure TCP Forwarder
Custom Neovim Plugins
Datacenter
Embedded Replicated Key Value Store (WIP)
Glock
LLM Obfuscater
My Dev setup
Pending Ideas
Simple Remote FS
vcat
claw
README
wiki
wiki
concepts
agent-memory
ai-access-inequality
ai-alignment
ai-character-and-personality
autonomous-research
coding-agents
cognitive-debt
compounding-knowledge
context-engineering
continual-learning
llm-as-knowledge-worker
mechanistic-interpretability
memex
personal-knowledge-management
prefill-decode-disaggregation
rag-vs-compiled-knowledge
reinforcement-learning-infrastructure
reward-hacking
scaling-laws-and-compute
sycophancy-problem
synthetic-data
test-time-compute
entities
alex-xi
andrej-karpathy
anthropic
cursor
fireworks-ai
letta
lilian-weng
moonshot-ai
nathan-lambert
openai
simon-willison
log
schema
sources
2026-and-beyond-tu
a-visual-guide-quantization
abundant-intelligence-altman
advancing-search-augmented-lms
ai-2025-reading-list-survey
ai-2027-stumbling-agents
ai-600b-question
ai-agent-security-windows95
ai-does-progress-have-speed-limit
ai-founders-bitter-lesson
ai-native-interview-sierra
ai-products-probabilistic-era
ai-vision-what-2030-looks-like
alex-wiki-discussion-2026-04-06
amanda-askell-ai-consciousness
ampcode-how-to-build-agent
anatomy-modern-finetuning-api
anthropic-ai-resistant-evals
anthropic-claude-character
anthropic-context-engineering
anthropic-demystifying-evals
anthropic-effective-harnesses
anthropic-emotion-concepts
anthropic-glasswing
anthropic-postmortem-infrastructure
anthropic-reward-hacking-sabotage
anthropic-situational-awareness
april-rl-active-partial-rollouts
arcee-trinity-large
attention-wrong-norm
atuin-runbooks-that-run
autoevolver-coding-agents
avatarl-pure-rl-training
behaviourisms-revenge
big-companies-vs-startups
brainstore-architecture
building-background-agent-ramp
building-generative-ai-platform
building-llm-apps-production-chip
building-state-of-art-agents-mercor
buildling-ai-native-products
bytebytego-how-claude-thinks
can-open-source-ai-introspect
cate-hall-how-to-be-agentic
chain-of-thought-serial-problems
chinchilla-compute-optimal-training
chinese-open-source-history
chip-huyen-agents
chroma-context-1-search-agent
claude-code-analysis
claude-code-inflection-point
claude-code-rl-reward-hacking
claude-reverse-engineered-memory
claude-system-prompt-highlights
closing-of-the-frontier
comparing-claude-chatgpt-memory
context-engineering-manus
context-infrastructure-grapeot
context-rot-chroma
continual-learning-problem
contra-dspy-gepa
cursor-bench-internal-eval
cursor-real-time-rl-composer
cursor-scaling-agents
cursor-secure-codebase-indexing
cursor-self-summarization
cursor-warp-decode
dario-adolescence-of-technology
dario-amodei-machines-of-loving-grace
dario-amodei-on-openai-military
databases-in-2025-pavlo
deadly-triad-rl-instability
deepseek-before-v4
deepseek-v4-before-release
demis-hassabis-agents-agi
differential-transformer
do-things-dont-scale
doc-to-lora-sakana
does-rl-incentivize-reasoning
dpo-reward-model
duolingo-reignited-growth
dwarkesh-ai-progress-dec2025
factory-signals-self-improving
failing-to-understand-exponential
feedback-flywheel
fields-notes-shipping-with-claude
fine-tuning-as-service
fireworks-frontier-rl-cheaper
first-fully-general-cam
fluid-representations-reasoning
founder-mode-pg
from-hierarchy-intelligence-sequoia
frontier-pretraining-tpu-gpt-oss
galaxy-brain-resistance
gemma-4-google
generative-user-models
goodharts-law-wikipedia
google-deepmind-deal
google-notebooklm-podcasts
gpu-glossary-modal
gpu-memory-snapshots-modal
gpu-mode-and-kernel-engineering
grpo-tricks-for-rl
hacking-discord-vercel
harness-engineering
hindsight-agent-memory-architecture
hindsight-vectorize-memory
hitchhikers-guide-llm-agent
hitting-1000-tokens-rtx5090
how-attention-residuals-work
how-copilot-makes-programmers-worse
how-to-build-an-agent
how-to-hire-chief-of-staff
hw-sw-codesign-matx-video
i-reverse-engineered-200-ai-startups
icl-rm-latent-compression-report
ilya-reading-list
implicitrm-unbiased-reward-modeling
improving-cursor-tab-online-rl
inside-ai-workflows-every
inside-cursor
jagged-ai-frontier-data
jason-wei-asymmetry-verification
jike-lovart-genspark-emochi-growth
karpathy-autoresearch
karpathy-deep-neural-nets-33-years
karpathy-llm-wiki-pattern
karpathy-llm-wiki
karpathy-recipe-training-neural-networks
keeping-20000-gpus-healthy
keith-rabois-building-ai-era
language-models-as-scaffolds
learnable-programming-bret-victor
learning-next-action-predictors
let-verify-step-by-step
letta-context-repositories
lilian-weng-reward-hacking
lilian-weng-transformer-inference
lilian-weng-why-we-think
linear-agent-interaction-guidelines
llm-daydreaming
llm-inference-economics
llm-judges-reward-models
llm-personalization-reward-factorization
llms-find-people
llms-not-getting-better
lora-without-regret
lovable-growth-story
machines-of-loving-grace-poem
making-rl-fast
mastra-observational-memory
mem-agent-persistent-memory-rl
memory-for-ai-biologically-inspired
mercury-edit-2
meta-harness-automatic-optimization
meta-harness-update-2026
metr-measuring-ai-long-tasks
mintlify-virtual-filesystem
mit-hyung-won-chung-incentivize
mitchell-hashimoto-ai-adoption
mixture-experts-architectures
nathan-lambert-job-market-llms
nathan-lambert-lossy-self-improvement
natural-language-agent-harnesses
neel-nanda-problems-for-fixing
network-effects-manual
next-llm-cost-saving
nik-pash-no-rag-for-agents
nous-hermes-agent
nova-act-as-service
on-deepseek-export-controls
on-policy-distillation
onecli-credential-vault
open-social-overreacted
open-weights-open-training
openai-chatgpt-images
openai-equipping-agents-skills
openai-goblins-personality
openai-real-world-tasks
openai-reverse-engineered
openai-scaling-postgresql
openclaw-rl-continuous-optimization
openrubrics-reward-modeling
polaris-post-training-recipe
policy-gradient-algorithms-lilian
post-training-101
predictions-continual-learning
prefill-as-a-service
quantization-from-ground-up
ramp-background-agent
recursive-language-models-2026
reflections-on-openai
remote-prompt-injection-gitlab
representation-engineering-mistral
reward-hacking-nvfp4
rl-collapse-training-inference
rl-environments-faq
rl-from-scratch-avatarl
rl-from-user-conversations
rl-gptreason-moment
rl-information-bandwidth
rl-is-new-fine-tuning
roko-basilisk
rubrics-as-rewards-paper
sam-altman-intelligence-age
self-improving-agent-dynamic-context
sequoia-hierarchy-to-intelligence
sft-equals-rl
shunyu-yao-second-half
simonwillison-agent-definition
simonwillison-slowing-down
skypilot-scaling-autoresearch
slack-managing-context-agents
slate-beyond-react-rlm
software-2-karpathy
spec-driven-development
stateful-agents-letta-tutorial
stop-pretending-ai-economy
supporting-mixtral-gpt-fast
swyx-agent-labs-thesis
swyx-ai-engineer-reading-list
synthetic-data-fine-tuning
synthetic-pretraining-vintagedata
textual-steering-vectors-multimodal
the-bitter-lesson
the-continual-learning-problem
there-are-no-new-ideas-ai
thoughts-on-evals
thunderkittens-gpu-kernels
tiktok-sorting-hat-eugene-wei
token-economy-bubble-update
token-economy-bubble
training-rlhf-for-voice
training-self-correct-rl
training-vlm-for-cua
turboquant-kv-cache-compression
tweet-100k-h100-datacenter-numbers
tweet-agent-output-as-compiler-output
tweet-agentcraft-rts-interface
tweet-ai-agent-armies-comparison
tweet-ai-funding-openai-recruiting
tweet-ai-microdramas-china-3b
tweet-ai-semiconductor-endgame
tweet-ai-startups-path-forward
tweet-ai-supply-chain-map
tweet-anthropic-distillation-attack
tweet-anti-moe-take
tweet-arc-agi-3-announcement
tweet-asymmetric-verification-law
tweet-attention-free-model-14b
tweet-auto-research-legal-agents
tweet-autoagent-self-optimizing
tweet-building-spectre
tweet-causal-jepa-icml
tweet-character-ai-pretraining-tricks
tweet-china-ai-labs-vibes
tweet-claude-code-sentiment-detection
tweet-claude-code-source-leaked
tweet-claude-code-spec-workflow
tweet-claude-soul-document
tweet-coding-workflow-shift
tweet-comprehensive-rl-overview-murphy
tweet-continualcode-online-learning
tweet-cuda-matmul-beats-cublas
tweet-curiosity-driven-jepa-rl
tweet-curiosity-reward-personalization
tweet-cursor-adoption-study
tweet-cursor-composer-rl-details
tweet-cursor-grep-instant
tweet-cursor-vs-cognition-search
tweet-data-center-ai-primer
tweet-data-efficiency-gpu-abuse
tweet-datacenter-economics-stanford
tweet-deepseek-v4-analysis
tweet-deepseek-v4-hbm
tweet-defining-continual-learning
tweet-deployed-intelligence-thesis
tweet-dr-grpo-interview
tweet-dwarkesh-frontier-training-lecture
tweet-eurekaclaw-research-agent
tweet-finding-right-altitude
tweet-fine-tuning-business
tweet-flashattention-most-innovative
tweet-fp16-vs-bf16-rl-training
tweet-future-interface-three-layers
tweet-gemma4-architecture
tweet-generative-meta-model-activations
tweet-gpu-failure-cost
tweet-group-representational-pos-encoding
tweet-grpo-group-normalization-problem
tweet-hotz-tier-list
tweet-hw-sw-codesign-prefill-decode
tweet-hyperagents-self-accelerating
tweet-iclr-scaling-rl-1t
tweet-information-consumption-breakdown
tweet-infra-guide-ai-agents
tweet-intelligence-per-picojoule
tweet-karpathy-1b-clean-data
tweet-kv-cache-nightmares
tweet-llm-invertibility
tweet-llm-knowledge-bases
tweet-megakernels-training
tweet-metr-gpt5-horizon
tweet-model-size-from-knowledge
tweet-moe-training-instability
tweet-on-policy-distillation-multi-turn
tweet-openreward-rl-environments
tweet-parallel-agent-context-search
tweet-physera-applied-intelligence
tweet-predictive-user-model
tweet-pretraining-kv-cache
tweet-pretraining-vs-posttraining
tweet-prime-intellect-democratizing
tweet-pufferlib-4-rl-training
tweet-ramp-ai-coworker
tweet-returns-intelligence-nonlinear
tweet-rl-entropy-collapse-tricks
tweet-rl-entry-guide-chinese
tweet-rl-env-api-design
tweet-rl-environment-startups-pavlov
tweet-rl-kl-divergence-first-principles
tweet-rl-vs-opd-vs-sft-survey
tweet-rlhf-infra-miles-docs
tweet-rubrics-as-rewards
tweet-scaling-axes-history
tweet-self-distillation-papers
tweet-semianalysis-claude-code-inflection
tweet-senior-engineers-accept-agent-output
tweet-sft-rl-kl-divergence
tweet-shopify-ai-baseline
tweet-shopify-qwen3-finetuned
tweet-silico-model-debugging
tweet-skills-vs-harnesses-agent-startups
tweet-sparse-attention-anns
tweet-subq-sub-quadratic-llm
tweet-surge-ai-data-labeling
tweet-systems-engineering-agentic
tweet-test-time-scaling-extent
tweet-tiny-rl-repo
tweet-transformer-circuits-mental-model
tweet-ttt-rl-stanford-nvidia
tweet-value-functions-for-llm-agents
tweet-vera-rubin-value-capture
tweet-version-control-ai-coding
tweet-vibe-coded-cuda-inference
tweet-xai-post-training-rl
tweet-xai-rl-for-atoms
understanding-sft-wolfe
vinegar-dumplings-sglang
vllm-triton-attention-backend
vmax-reinforcement-learning
wall-between-shannon-kolmogorov
we-dont-sell-saddles-here
we-interviewed-100-eng-teams
welcome-to-machine-agent-infra
what-mcp-not-needed
what-will-be-scarce
where-is-your-moat
why-ai-progress-invisible
why-bootstrapping-readwise
synthesis
coding-agent-paradox
rl-infrastructure-maturation
Home
❯
wiki
❯
sources
❯
rl-from-scratch-avatarl
rl-from-scratch-avatarl
Jul 06, 2026
1 min read
pure-rl
language-models
training
research
Pure RL Training of Language Models
Note: Already covered in
avatarl-pure-rl-training
.
Search
Search
Backlinks
No backlinks found
Explorer
Writings
Debugging Issues from Postgres JSONB
Engineering
AI Team Org Map (April 2026)
AI Team Org Map (April 2026).sync-conflict-20260406-161943-6HXDDG5
AI Team Org Map (April 2026).sync-conflict-20260406-162533-6HXDDG5
Framework for Technical Decisions
ML
GPU Reading Group
Reading Group Plan
Session 1
AI
day of session guide
email templates
exercise code files
gpu study session1 plan
host preparation
quick prep checklist
quick reference card
session backup plans
session materials overview
Agenda
Prep
Prereqs
Session 2
Prep
Notes On Finetuning
My experience with GPU vendors
On scaling an engineering stack (WIP)
Startup Engineering Growth Stages
To the interviewees
Mindset
Active Thinking and Processes (WIP)
Random
Startups
Capturing Value From Technology Diffusion (WIP)
On Serverless Providers (WIP)
On startup velocity (WIP)
Templates
Daily Note Template
Weekly Note Template
Weekend Projects
Bardock - a Monorepo Dockerfile management tool
Building a secure TCP Forwarder
Custom Neovim Plugins
Datacenter
Embedded Replicated Key Value Store (WIP)
Glock
LLM Obfuscater
My Dev setup
Pending Ideas
Simple Remote FS
vcat
claw
README
wiki
wiki
concepts
agent-memory
ai-access-inequality
ai-alignment
ai-character-and-personality
autonomous-research
coding-agents
cognitive-debt
compounding-knowledge
context-engineering
continual-learning
llm-as-knowledge-worker
mechanistic-interpretability
memex
personal-knowledge-management
prefill-decode-disaggregation
rag-vs-compiled-knowledge
reinforcement-learning-infrastructure
reward-hacking
scaling-laws-and-compute
sycophancy-problem
synthetic-data
test-time-compute
entities
alex-xi
andrej-karpathy
anthropic
cursor
fireworks-ai
letta
lilian-weng
moonshot-ai
nathan-lambert
openai
simon-willison
log
schema
sources
2026-and-beyond-tu
a-visual-guide-quantization
abundant-intelligence-altman
advancing-search-augmented-lms
ai-2025-reading-list-survey
ai-2027-stumbling-agents
ai-600b-question
ai-agent-security-windows95
ai-does-progress-have-speed-limit
ai-founders-bitter-lesson
ai-native-interview-sierra
ai-products-probabilistic-era
ai-vision-what-2030-looks-like
alex-wiki-discussion-2026-04-06
amanda-askell-ai-consciousness
ampcode-how-to-build-agent
anatomy-modern-finetuning-api
anthropic-ai-resistant-evals
anthropic-claude-character
anthropic-context-engineering
anthropic-demystifying-evals
anthropic-effective-harnesses
anthropic-emotion-concepts
anthropic-glasswing
anthropic-postmortem-infrastructure
anthropic-reward-hacking-sabotage
anthropic-situational-awareness
april-rl-active-partial-rollouts
arcee-trinity-large
attention-wrong-norm
atuin-runbooks-that-run
autoevolver-coding-agents
avatarl-pure-rl-training
behaviourisms-revenge
big-companies-vs-startups
brainstore-architecture
building-background-agent-ramp
building-generative-ai-platform
building-llm-apps-production-chip
building-state-of-art-agents-mercor
buildling-ai-native-products
bytebytego-how-claude-thinks
can-open-source-ai-introspect
cate-hall-how-to-be-agentic
chain-of-thought-serial-problems
chinchilla-compute-optimal-training
chinese-open-source-history
chip-huyen-agents
chroma-context-1-search-agent
claude-code-analysis
claude-code-inflection-point
claude-code-rl-reward-hacking
claude-reverse-engineered-memory
claude-system-prompt-highlights
closing-of-the-frontier
comparing-claude-chatgpt-memory
context-engineering-manus
context-infrastructure-grapeot
context-rot-chroma
continual-learning-problem
contra-dspy-gepa
cursor-bench-internal-eval
cursor-real-time-rl-composer
cursor-scaling-agents
cursor-secure-codebase-indexing
cursor-self-summarization
cursor-warp-decode
dario-adolescence-of-technology
dario-amodei-machines-of-loving-grace
dario-amodei-on-openai-military
databases-in-2025-pavlo
deadly-triad-rl-instability
deepseek-before-v4
deepseek-v4-before-release
demis-hassabis-agents-agi
differential-transformer
do-things-dont-scale
doc-to-lora-sakana
does-rl-incentivize-reasoning
dpo-reward-model
duolingo-reignited-growth
dwarkesh-ai-progress-dec2025
factory-signals-self-improving
failing-to-understand-exponential
feedback-flywheel
fields-notes-shipping-with-claude
fine-tuning-as-service
fireworks-frontier-rl-cheaper
first-fully-general-cam
fluid-representations-reasoning
founder-mode-pg
from-hierarchy-intelligence-sequoia
frontier-pretraining-tpu-gpt-oss
galaxy-brain-resistance
gemma-4-google
generative-user-models
goodharts-law-wikipedia
google-deepmind-deal
google-notebooklm-podcasts
gpu-glossary-modal
gpu-memory-snapshots-modal
gpu-mode-and-kernel-engineering
grpo-tricks-for-rl
hacking-discord-vercel
harness-engineering
hindsight-agent-memory-architecture
hindsight-vectorize-memory
hitchhikers-guide-llm-agent
hitting-1000-tokens-rtx5090
how-attention-residuals-work
how-copilot-makes-programmers-worse
how-to-build-an-agent
how-to-hire-chief-of-staff
hw-sw-codesign-matx-video
i-reverse-engineered-200-ai-startups
icl-rm-latent-compression-report
ilya-reading-list
implicitrm-unbiased-reward-modeling
improving-cursor-tab-online-rl
inside-ai-workflows-every
inside-cursor
jagged-ai-frontier-data
jason-wei-asymmetry-verification
jike-lovart-genspark-emochi-growth
karpathy-autoresearch
karpathy-deep-neural-nets-33-years
karpathy-llm-wiki-pattern
karpathy-llm-wiki
karpathy-recipe-training-neural-networks
keeping-20000-gpus-healthy
keith-rabois-building-ai-era
language-models-as-scaffolds
learnable-programming-bret-victor
learning-next-action-predictors
let-verify-step-by-step
letta-context-repositories
lilian-weng-reward-hacking
lilian-weng-transformer-inference
lilian-weng-why-we-think
linear-agent-interaction-guidelines
llm-daydreaming
llm-inference-economics
llm-judges-reward-models
llm-personalization-reward-factorization
llms-find-people
llms-not-getting-better
lora-without-regret
lovable-growth-story
machines-of-loving-grace-poem
making-rl-fast
mastra-observational-memory
mem-agent-persistent-memory-rl
memory-for-ai-biologically-inspired
mercury-edit-2
meta-harness-automatic-optimization
meta-harness-update-2026
metr-measuring-ai-long-tasks
mintlify-virtual-filesystem
mit-hyung-won-chung-incentivize
mitchell-hashimoto-ai-adoption
mixture-experts-architectures
nathan-lambert-job-market-llms
nathan-lambert-lossy-self-improvement
natural-language-agent-harnesses
neel-nanda-problems-for-fixing
network-effects-manual
next-llm-cost-saving
nik-pash-no-rag-for-agents
nous-hermes-agent
nova-act-as-service
on-deepseek-export-controls
on-policy-distillation
onecli-credential-vault
open-social-overreacted
open-weights-open-training
openai-chatgpt-images
openai-equipping-agents-skills
openai-goblins-personality
openai-real-world-tasks
openai-reverse-engineered
openai-scaling-postgresql
openclaw-rl-continuous-optimization
openrubrics-reward-modeling
polaris-post-training-recipe
policy-gradient-algorithms-lilian
post-training-101
predictions-continual-learning
prefill-as-a-service
quantization-from-ground-up
ramp-background-agent
recursive-language-models-2026
reflections-on-openai
remote-prompt-injection-gitlab
representation-engineering-mistral
reward-hacking-nvfp4
rl-collapse-training-inference
rl-environments-faq
rl-from-scratch-avatarl
rl-from-user-conversations
rl-gptreason-moment
rl-information-bandwidth
rl-is-new-fine-tuning
roko-basilisk
rubrics-as-rewards-paper
sam-altman-intelligence-age
self-improving-agent-dynamic-context
sequoia-hierarchy-to-intelligence
sft-equals-rl
shunyu-yao-second-half
simonwillison-agent-definition
simonwillison-slowing-down
skypilot-scaling-autoresearch
slack-managing-context-agents
slate-beyond-react-rlm
software-2-karpathy
spec-driven-development
stateful-agents-letta-tutorial
stop-pretending-ai-economy
supporting-mixtral-gpt-fast
swyx-agent-labs-thesis
swyx-ai-engineer-reading-list
synthetic-data-fine-tuning
synthetic-pretraining-vintagedata
textual-steering-vectors-multimodal
the-bitter-lesson
the-continual-learning-problem
there-are-no-new-ideas-ai
thoughts-on-evals
thunderkittens-gpu-kernels
tiktok-sorting-hat-eugene-wei
token-economy-bubble-update
token-economy-bubble
training-rlhf-for-voice
training-self-correct-rl
training-vlm-for-cua
turboquant-kv-cache-compression
tweet-100k-h100-datacenter-numbers
tweet-agent-output-as-compiler-output
tweet-agentcraft-rts-interface
tweet-ai-agent-armies-comparison
tweet-ai-funding-openai-recruiting
tweet-ai-microdramas-china-3b
tweet-ai-semiconductor-endgame
tweet-ai-startups-path-forward
tweet-ai-supply-chain-map
tweet-anthropic-distillation-attack
tweet-anti-moe-take
tweet-arc-agi-3-announcement
tweet-asymmetric-verification-law
tweet-attention-free-model-14b
tweet-auto-research-legal-agents
tweet-autoagent-self-optimizing
tweet-building-spectre
tweet-causal-jepa-icml
tweet-character-ai-pretraining-tricks
tweet-china-ai-labs-vibes
tweet-claude-code-sentiment-detection
tweet-claude-code-source-leaked
tweet-claude-code-spec-workflow
tweet-claude-soul-document
tweet-coding-workflow-shift
tweet-comprehensive-rl-overview-murphy
tweet-continualcode-online-learning
tweet-cuda-matmul-beats-cublas
tweet-curiosity-driven-jepa-rl
tweet-curiosity-reward-personalization
tweet-cursor-adoption-study
tweet-cursor-composer-rl-details
tweet-cursor-grep-instant
tweet-cursor-vs-cognition-search
tweet-data-center-ai-primer
tweet-data-efficiency-gpu-abuse
tweet-datacenter-economics-stanford
tweet-deepseek-v4-analysis
tweet-deepseek-v4-hbm
tweet-defining-continual-learning
tweet-deployed-intelligence-thesis
tweet-dr-grpo-interview
tweet-dwarkesh-frontier-training-lecture
tweet-eurekaclaw-research-agent
tweet-finding-right-altitude
tweet-fine-tuning-business
tweet-flashattention-most-innovative
tweet-fp16-vs-bf16-rl-training
tweet-future-interface-three-layers
tweet-gemma4-architecture
tweet-generative-meta-model-activations
tweet-gpu-failure-cost
tweet-group-representational-pos-encoding
tweet-grpo-group-normalization-problem
tweet-hotz-tier-list
tweet-hw-sw-codesign-prefill-decode
tweet-hyperagents-self-accelerating
tweet-iclr-scaling-rl-1t
tweet-information-consumption-breakdown
tweet-infra-guide-ai-agents
tweet-intelligence-per-picojoule
tweet-karpathy-1b-clean-data
tweet-kv-cache-nightmares
tweet-llm-invertibility
tweet-llm-knowledge-bases
tweet-megakernels-training
tweet-metr-gpt5-horizon
tweet-model-size-from-knowledge
tweet-moe-training-instability
tweet-on-policy-distillation-multi-turn
tweet-openreward-rl-environments
tweet-parallel-agent-context-search
tweet-physera-applied-intelligence
tweet-predictive-user-model
tweet-pretraining-kv-cache
tweet-pretraining-vs-posttraining
tweet-prime-intellect-democratizing
tweet-pufferlib-4-rl-training
tweet-ramp-ai-coworker
tweet-returns-intelligence-nonlinear
tweet-rl-entropy-collapse-tricks
tweet-rl-entry-guide-chinese
tweet-rl-env-api-design
tweet-rl-environment-startups-pavlov
tweet-rl-kl-divergence-first-principles
tweet-rl-vs-opd-vs-sft-survey
tweet-rlhf-infra-miles-docs
tweet-rubrics-as-rewards
tweet-scaling-axes-history
tweet-self-distillation-papers
tweet-semianalysis-claude-code-inflection
tweet-senior-engineers-accept-agent-output
tweet-sft-rl-kl-divergence
tweet-shopify-ai-baseline
tweet-shopify-qwen3-finetuned
tweet-silico-model-debugging
tweet-skills-vs-harnesses-agent-startups
tweet-sparse-attention-anns
tweet-subq-sub-quadratic-llm
tweet-surge-ai-data-labeling
tweet-systems-engineering-agentic
tweet-test-time-scaling-extent
tweet-tiny-rl-repo
tweet-transformer-circuits-mental-model
tweet-ttt-rl-stanford-nvidia
tweet-value-functions-for-llm-agents
tweet-vera-rubin-value-capture
tweet-version-control-ai-coding
tweet-vibe-coded-cuda-inference
tweet-xai-post-training-rl
tweet-xai-rl-for-atoms
understanding-sft-wolfe
vinegar-dumplings-sglang
vllm-triton-attention-backend
vmax-reinforcement-learning
wall-between-shannon-kolmogorov
we-dont-sell-saddles-here
we-interviewed-100-eng-teams
welcome-to-machine-agent-infra
what-mcp-not-needed
what-will-be-scarce
where-is-your-moat
why-ai-progress-invisible
why-bootstrapping-readwise
synthesis
coding-agent-paradox
rl-infrastructure-maturation