Dr Clinton Mielke (Cosmo) retweeted
in all my years of doing neuroscience, I never thought we would end up here
Dr Clinton Mielke (Cosmo) retweeted
Replying to @AnthropicAI
Dr Clinton Mielke (Cosmo) retweeted
Replying to @BananaBatmans @zerophases
My good fellow, we are all of us primates here.
Dr Clinton Mielke (Cosmo) retweeted
As Larry Ellison tops Elon Musk as world's richest person, remember that Ellison has supported basic research on aging.
The Ellison Medical Foundation, e.g., funded our work with @geochurch on naked mole rats genomics.
Ellison also recently founded the impressive @eit_cx
Larry Ellison tops Elon Musk as world's richest person bbc.in/4gfY2qW
Dr Clinton Mielke (Cosmo) retweeted
Pfizer discovered the antifungal fluconazole - a cure for candida infections and deadly fungal meningitis
Pfizer discovered the tetracycline antibiotic which cures a wide range of infections
Pfizer massively scaled up penicillin productio, CURING millions from infection
🤡
This quoted post is unavailable.
Dr Clinton Mielke (Cosmo) retweeted
Replying to @patriottakes
Just the Russians doing a test run using the White House windows
Dr Clinton Mielke (Cosmo) retweeted
She dumped me last night.
Not because I don't listen.
Not because I'm always on my phone.
Not even because I forgot our anniversary (twice).
But because,
in her exact words:
"You only pay attention to the parts of what I say that you think are important."
I stared at her for a moment and realized...
She just perfectly described the attention mechanism in transformers.
Turns out I wasn't being a bad boyfriend. I was being mathematically optimal.
See, in conversations (and transformers), you don't give equal weight to every word. Some words matter more for understanding context. Attention figures out exactly HOW important each word should be.
Here's the beautiful math:
Attention(Q, K, V) = softmax(QK^T / √d_k)V
Breaking it down:
Q (Query): "What am I looking for?"
K (Key): "What info is available?"
V (Value): "What is that info?"
d_k: Key dimension (for scaling)
Think library analogy:
You have a question (Query). Books have titles (Keys) and content (Values). Attention finds which books are most relevant.
Step-by-step with "The cat sat on the mat":
Step 1: Create Q, K, VEach word → three vectors via learned matrices W_Q, W_K, W_V
For "cat":
Query: "What should I attend to when processing 'cat'?"
Key: "I am 'cat'"
Value: "Here's cat info"
Step 2: Calculate scoresQK^T = how much each word should attend to others
Processing "sat"? High similarity with "cat" (cats sit) and "mat" (where sitting happens).
Step 3: Scale by √d_kPrevents dot products from getting too large, keeps softmax balanced.
Step 4: SoftmaxConverts scores to probabilities:
"cat": 0.4 (subject)
"sat": 0.3 (action)
"mat": 0.2 (location)
"on": 0.1 (preposition)
"the": 0.1 (article)
Step 5: Weight valuesMultiply each word's value by attention weight, sum up. Now "sat" knows it's most related to "cat" and "mat".
Multi-Head Magic:Transformers do this multiple times in parallel:
Head 1: Subject-verb relationships
Head 2: Spatial ("on", "in", "under")
Head 3: Temporal ("before", "after")
Head 4: Semantic similarity
Each head learns different relationship types.
Why This Changed Everything:
Before: RNNs = reading with flashlight (one word at a time, forget the beginning)
After: Attention = floodlights on entire sentence with dimmer switches
This is why ChatGPT can:
Remember 50 messages ago
Know "it" refers to something specific
Understand "bank" = money vs river based on context
The Kicker:Models learn these patterns from data alone. Nobody programmed grammar rules. It figured out language structure just by predicting next words.
Attention is how AI learned to read between the lines.
Just like my therapist helped me understand my focus patterns, maybe understanding transformers helps us see how we decide what matters.
Now if only I could implement multi-head attention in dating... 🤖
Still waiting for "scaled dot-product listening" to be invented.
Dr Clinton Mielke (Cosmo) retweeted
Replying to @DNIGabbard @realDonaldTrump
Bottom line: Did the Obama Administration manufacture false intelligence to implicate Russian interference in the 2016 election?
No.
The documents show a methodical, multi-agency process, all concluding Russia was behind the DNC hacks and intended to influence perceptions surrounding the election. Internal emails show open debate, requested edits for accuracy, and even pushback from the FBI urging softer language where intelligence was not definitive. This shows a clear intent to avoid politicization, not to fabricate Intel.
The Obama Administration did not generate intelligence from thin air. It commissioned a community assessment in response to mounting cyber activity and growing concerns from state officials. The reports repeatedly emphasize the limited impact Russia could have on the actual vote, while also noting the psychological and political effects of leaks and disinformation.
These documents don't show the Obama Administration cooked up fake intel to blame Russia. They reflect a serious, cautious, and collaborative process across agencies, with multiple layers of review and disagreement. What they found was cyber interference and information warfare designed to mess with our heads, not our ballots.
Russia got inside the DNC, released damaging info, and aimed to create chaos. That’s what the intel says, nothing more, nothing less. And they don’t link Trump’s team to it in these pages, nor do they try to.
These declassified intelligence documents show internal communications and assessments among senior officials across the IC, with the purpose of evaluating cyber threats related to the 2016 presidential election.
The findings can be broken down like this:
- Capability Assessment: The IC judged with moderate confidence that no foreign adversary, including Russia, had the capability to carry out widespread, covert cyber operations that could alter the vote count or outcome of the 2016 U.S. election.
- Russian Activity: Russia was identified as the most capable and likely actor interested in undermining the election. High-confidence assessments attributed the hacks of the DNC and DCCC to Russian intelligence.
- Intent vs. Impact: The most likely threat, according to the documents, was not manipulation of vote totals but cyber activity designed to sow distrust and cast doubt on the legitimacy of the electoral process. That includes theft and release of information, cyber intrusions into voter databases, and media disinformation.
- Local Intrusions: Voter registration systems in Illinois and Arizona were breached, but no evidence showed tampering with actual votes. 20 other states experienced probing or vulnerability scans, most traced to Russian-operated servers. Still, none of the observed activity suggested an actual alteration of results.
- No Evidence of Manipulated Vote Counts: Multiple safeguards, decentralized systems, and lack of evidence led the IC to conclude with high confidence that no vote manipulation occurred through cyber means.
- Trump Campaign: No direct link between the Trump campaign and Russia is mentioned in the documents. The documents focus strictly on cyber activities, infrastructure vulnerabilities, foreign intentions, and information operations.
This declassification vindicates the Obama administration and makes DNI Gabbard look foolish.
Dr Clinton Mielke (Cosmo) retweeted
I tried... 😅
Replying to @sokrypton
Make an unfolding video and reverse it. Like this: teapawt.org/andrewwhite01/status/1…
Dr Clinton Mielke (Cosmo) retweeted
Tucker: How many people live in Iran?
Cruz: I don’t know the population
Tucker: You don’t know the population of the country you seek to topple?
Cruz: How many people live in Iran?
Tucker: 92 million. How could you not know that?
Dr Clinton Mielke (Cosmo) retweeted
Please join us in Palo Alto this Saturday June 21 @ 10a for a 𝐜𝐞𝐥𝐞𝐛𝐫𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐃𝐫. 𝐀𝐭𝐮𝐥 𝐁𝐮𝐭𝐭𝐞’𝐬 𝐥𝐢𝐟𝐞 co-hosted by UCSF, UC Health, Stanford, and Harvard
stanfordmedicine.qualtrics.c…
Dr Clinton Mielke (Cosmo) retweeted
Replying to @EM_RESUS
90 minutes of CPR with a Lucas device. He opened his eyes and looked around the room while in VFib with compressions. Went home a couple of weeks later. Now back at work. It was glorious.
Dr Clinton Mielke (Cosmo) retweeted
Replying to @EM_RESUS
A perfectly healthy @DGlaucomflecken went to bed one night, and his heart arrested before morning. I did 10 minutes of CPR before paramedics arrived, and he survived physiologically and neurologically intact. The 911 dispatcher who instructed me to start CPR is my hero.
Dr Clinton Mielke (Cosmo) retweeted
Hi, ER Doc here.
Repost if you’ve ever done CPR on someone and they survived.
Share your story in the thread.
Dr Clinton Mielke (Cosmo) retweeted
Scoop: The databases powering doge. gov are insecure, and people outside the government have already pushed their own updates to the site to prove it:
404media.co/anyone-can-push-…
Dr Clinton Mielke (Cosmo) retweeted
I have hesitated to share this document I received earlier this year, mostly for my own sanity, but on this, the anniversary of my father's death, I feel obliged to share it in hope that it leads to someone somewhere having a better fate than him.
Dr Clinton Mielke (Cosmo) retweeted
Do I have bad news for you…
Dr Clinton Mielke (Cosmo) retweeted
Listen, if you want universal healthcare – the only way forward is obligatory universal ozempic. Factories in every state pumping it out, cheaper than gallons of water at Walmart. We are not going to fix the food. Your pet theories. We've had 50 years. This is it. It's _over_.
Dr Clinton Mielke (Cosmo) retweeted
This is my favorite cell embedding paper! SATURN learns universal cell embeddings by coupling gene expression with pLM embeddings for cross-species integration of functionally-related genes (macrogenes). 🌟 Amazing for atlas integration!
Paper: nature.com/articles/s41592-0…
Code: github.com/snap-stanford/sat…



































