Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
I captured this video in my lab to show the power of medicines called T-cell engagers.
Left: T cells move in and out of a cancer tumoroid, largely ignoring it.
Right: Add a T-cell engager, and the T cells attack. Once they recognize their targets, they tear through the tumor.
Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
The race for AGI
Script: Sherpa by Pocket FM
Video: Seedance 2.5
🤖 Made with AI
Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
Biopharma leaders - are you developing (or competing against) T-cell engagers?
We analyzed 750+ TCE programs across 450+ developers, including hundreds of early-stage Chinese programs to determine what's trending and where the space is headed next.
While TCEs are approved in oncology, nearly all of the excitement is driven by the potential in autoimmune disease.
This makes sense - TCEs have largely succeeded when they target "clean", lineage-restricted cells, which is why they've worked well in B-cell malignancies or multiple myeloma but struggled against shared antigens in solid tumors. Autoimmune disease ports a lot of that underlying biology that made them successful, while opening up massive, chronic markets.
The rules of engagement are different here though - the winner won't just be the deepest depleter.
1st gen autoimmune TCEs were repurposed from oncology, but the disease profile is so different: oncology tolerates substantial tox in exchange for tumor killing, but autoimmune patients have a non-fatal chronic disease, and thus much less tolerance for CRS or serious infections.
This is why the engineering around the molecule is so critical (a theme we see playing out across the industry right now). About half the entire TCE pipeline already includes engineering beyond a CD3 x TAA bispecific, with distinct levers emerging for TCEs designed specifically for autoimmune disease:
• Masking or conditional activation
• Dual B-cell targeting
• 2+1 avidity formats
• CD3 affinity attenuation or immune tuning
• SubQ administration / dosing optimizations
The number of levers available to developers now to design new TCE constructs + the wide array of potential autoimmune diseases with sub-standard SOC will lead to market fragmentation.
The optimal product profile for refractory SLE won't necessarily be the same one for RA. The underlying biology, current treatment paradigm, patient burden and specific manifestations (heme, neurologic, joint, etc.) will create different requirements around depletion depth, safety, durability & dosing.
And Pharma's commercial model is closely tied to building diseases franchises managed by similar specialists. In a more fragmented market, this creates an advantage for portfolios of closely related assets each targeting specific segments of the treatment sequence.
(Hint for biotechs to sharpen BD: map where your program complements Pharma's existing franchises or fills a gap in their treatment ladder)
Of course, that's not to say these drugs won't be huge. A well-designed TCE could own the messy middle: refractory patients who need deeper depletion than conventional biologics, but for whom CAR-T level risk is disproportionate. And as safety gets proven, winning assets will go earlier.
There's a lot more in the report, covering:
• China (of course)
• Emerging strategies for solid tumors
• Review of clinical readouts
If you'd like it, just comment "TCE" below and we'll send it over!
I Numbered 18 Clips. Codex Edited This Entire AI Music Video | தனித்த காதல் youtu.be/6t7SzUkIke8?si=tg8H… via @YouTube
I just published Do We Still Need Human Scientists? medium.com/@shibi76/do-we-st…
Do not take the annual plan for @ElevenLabs because once you do that, you cannot downgrade, upgrade, or switch to a monthly plan
Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
🚀 Huge release for watermarks-remover v0.7.0:
- SynthId-text removal is now combined with a humanizer
- Beta black-box watermark-stealing module to target and optimize against vendor-specific (Anthropic, OpenAI, Google...) watermarks
- Improving i18n support for morphologically rich / inflectional languages (Polish, Finnish, Turkish, etc.)
Thanks a lot to all the contributors and maintainers (@poorvith_mp @bentalay), this release is really a team effort 🎉
github.com/guillaumemeyer/wa…
Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
At OpenAI, it's felt like our researchers & engineers have been living in the future with Codex.
Today, we're bringing that same experience to life science researchers with the Rosalind Workbench.
Claude Designed Proteins That Actually Worked 🧬 youtu.be/WZnHIvN4jQo?si=T2z6… via @YouTube
Claude code has very limited computer-use features. It is not able to see the actual user interface of the apps in development:
- any dev server
- any local electron
- any local development app on the desktop
Claude code is useless for UI UX. It can check the code only @claudeai
Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
Your life insurance company is betting your premium on AI data centers. Here is what happens if that bet goes wrong.
First, understand the structure.
Building one AI data center costs tens of billions.
If Meta borrowed $30 billion directly, it shows up as massive debt on their balance sheet. Stock drops. Credit rating takes a hit. Investors panic.
So they created a workaround called an SPV. A Special Purpose Vehicle.
A separate shell company that exists for one purpose: own the data center and carry the debt.
Meta keeps roughly 20% ownership. A private credit partner like Blue Owl takes 80%. The SPV borrows the money.
Builds the data center. Then leases it back to Meta on a 15 to 20 year triple net lease. Meta pays rent instead of carrying debt.
On paper, Meta's balance sheet looks clean. The $30 billion in debt lives inside the SPV. Not on Meta's books.
Here is where your money enters.
Who lends to these SPVs? Insurance companies and pension funds. Apollo lent Meta $29 billion through Athene, its life insurance arm. Athene gets its capital from everyday people paying monthly premiums.
Canada Pension Plan put over $10 billion into AI data centers in one year. Retirement money of every working Canadian.
Because the lease is backed by Meta with a strong credit rating, these bonds get rated A+. Looks safe.
Now follow the chain if AI demand slows down.
Tech companies no longer need as many data centers. They renegotiate leases or walk away. The SPV stops generating revenue.
The A+ rated bonds crash in value because the assumptions behind them no longer hold.
$800 billion of US insurer money is locked in private credit and infrastructure debt. These are not stocks.
You cannot sell a bond backed by a half empty data center full of outdated GPUs. Nobody is buying.
But people still file claims. Retirees still need pensions. The insurance company needs cash. Its money is locked in assets it cannot sell.
And there is no safety net. Banks have FDIC. Insurance is regulated state by state. State systems can rescue one failing insurer. Not ten failing at the same time from the same shock.
The hardware problem makes it worse. These loans are modelled like 30 year real estate. But GPUs inside depreciate in 5 years.
Nvidia ships new chips every 18 months. The collateral loses value faster than the models assume.
If AI keeps growing, everything gets paid. System works.
If it does not, the people left holding the bag are not tech investors.
They are retirees and policyholders who never knew their safest money was inside an AI data center behind an SPV they have never heard of.
ChatGPT extension in Microsoft Edge browser is broken. Is it only for me or others are seeing it too? It works flawlessly in Chrome. Maybe there is a conflict with copilot or bing...
Dr. Shibichakravarthy Kannan MBBS, PhD retweeted
We’ve written an FAQ to answer some of the questions we've received about watermarking.
In summary:
• We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking;
• Our watermarking method doesn’t have any practical impact on the quality or content of Claude’s outputs;
• The difference between watermarked and un-watermarked text will not be distinguishable to readers;
• Nothing is added to the text and there are no hidden characters;
• Watermarking doesn’t require extra tokens, and will not be more expensive;
• Watermarks can’t be traced to a specific person, organization, or chat.
Read more: anthropic.com/news/claude-te…
Why can’t Gemini Spark organize Google Photos? It’s a privacy boundary by design.
Direct access is restricted to protect sensitive media. For AI curation, use the built-in Ask Photos feature in the Google Photos mobile app (Android/iOS). 📸📱 #GooglePhotos














